# Invenire — full site content > Founder-led AI engineering studio. We help teams turn AI prototypes > into production systems — agents, retrieval, vision, evals, telemetry, > and the engineering glue. Founded and run by Arve. Site: https://invenire.com.au/ Email: hello@invenire.com.au Based: Gold Coast, Australia (AU · GMT+10) This file concatenates every page on the site in markdown form so LLMs can ingest the whole site in a single fetch. The short curated index is at https://invenire.com.au/llms.txt. --- ## Home URL: https://invenire.com.au/ Pill: frontier · production · founder-led ### Hero **Frontier AI for the part after the demo.** I’m Arve. I build specialist models and production AI systems: image recognition, safety classification, multilingual speech, medical audio, and agents with a useful job to do. CTAs: Discuss an engagement (https://invenire.com.au/contact/) · Explore consulting (https://invenire.com.au/services/) Proof: Live now — Analytics Agent in the Shopify App Store. ### Credibility strip National-scale research infrastructure. Live AI products. Confidential agentic systems. One or two clients at a time. ### Stat strip - shipping_since: 2005 - production_ai_systems: 6+ - engagements_at_once: one or two - based_in: AU · GMT+10 ### What Invenire is Invenire is a founder-led AI engineering studio based on the Gold Coast, Australia. I design and ship agentic systems, retrieval and vision pipelines, evaluation infrastructure, analytics products, and AI-native engineering workflows for teams that need AI to work outside the demo. The models move quickly. I stay close to what is changing, test what matters, and bring clients the parts that hold up under production constraints. ### Flagship product: Analytics Agent **Eyebrow:** flagship · live · production **Link:** https://analytics-agent.app Six specialist agents read a Shopify merchant's GA4, Search Console and store data every day and write three to five plain-English findings. No dashboard. It is the live proof point — every architecture pattern, every eval technique, every retrieval choice I recommend on a call has already been pressure-tested here. - **mission_briefs** — Six agents, one router, daily brief. Rewritten three times before the briefs stopped reading like LLM slop. - **anomaly_detection** — Fifteen-minute checks against a rolling baseline. Catches the 2am spike before anyone logs in. - **ga4_audit** — Eight checks, a 0–100 score, one-click fixes. The fixes are the hard half — and the reason this exists. ### Services (home summary) **Four ways to bring in a senior AI engineer for the work that has to hold up.** - **01 / AI strategy and architecture.** Decide what to build, what to buy, and what to ignore. I help teams find the AI opportunities that are technically possible, commercially useful, and worth the operational risk. - **02 / Production AI systems.** Agents, retrieval, vision, and multi-modal workflows designed to run beyond the demo. Built with evals, telemetry, failure modes, and human review where they belong. - **03 / Evals and observability.** Golden datasets, regression tests, traces, cost visibility, latency tracking, and quality gates. Trust becomes something you can measure. - **04 / AI-native engineering.** Help your team adopt Cursor, Claude Code, agent workflows, eval-first development, and AI-assisted delivery without turning the codebase into an unreviewable mess. (Detailed write-ups: https://invenire.com.au/services/) ### Models trained to recognise what matters. I’ve trained image, text, classification, and audio models. Bring the examples your current system struggles with. That’s usually where the interesting work starts. - [Image recognition](https://invenire.com.au/services/image-recognition/) - [Safety classification](https://invenire.com.au/services/safety-classification/) - [Speech & medical audio](https://invenire.com.au/services/speech-medical-audio/) - [Legal & R&D review](https://invenire.com.au/services/legal-research-review/) → Explore custom models and local AI: https://invenire.com.au/services/model-development/ ### Operating principles (compact strip) currency · evidence · reliability · focus — full text on https://invenire.com.au/about/ ### Selected work (home summary) **Production systems. Not slides.** - **Analytics Agent** (SHOPIFY · LIVE · 2025–) — Six-agent system delivering daily Mission Briefs, fifteen-minute anomaly detection and a GA4 audit with one-click fixes to Shopify merchants. Router over specialists, LLM-as-judge evals wired into CI, full cost and latency telemetry per tool call. Live in the App Store. (https://analytics-agent.app) - **Plant** (SOLO-BUILT · SELF-HOSTING · 2025–) — Seven services, five languages, one operator. An autonomous engineering platform exploring how one senior operator can coordinate AI agents, externalised state, objective verification, and multi-service software delivery. The R&D layer behind the AI-native engineering practice. (https://invenire.com.au/work/plant/) - **EcoCommons** (NATIONAL · GRIFFITH · 2020–2023) — Australia's national platform for ecological and environmental modelling. Nine partner institutions, a decade of legacy systems re-engineered into one reproducible-by-design modelling platform. (https://invenire.com.au/work/ecocommons/) (All five case studies: https://invenire.com.au/work/) ### Founder (home summary) **Hi, I’m Arve.** PRINCIPAL_AI_ENGINEER · FOUNDER I’m the founder of Invenire. I’ve spent twenty years shipping software, from national research infrastructure to Shopify apps. These days I spend most of my time working out how to make AI useful when real people, data, and deadlines are involved. Invenire takes on a small number of engagements at a time, usually where architecture, implementation, and judgement all matter. Contact: hello@invenire.com.au · linkedin.com/in/arvesolland ### Closing CTA **Have a system that has to actually work?** Thirty minutes. Bring the architecture diagram or the screenshot of the bug — whichever is closer to where you are stuck. I will tell you what I would build, what I would not, and whether I am the right person for it. If I am not, I will say so. → Discuss an engagement: https://invenire.com.au/contact/ --- ## Services URL: https://invenire.com.au/services/ ### Hi, I’m Arve. Let’s build something useful. I train specialist models and build the software around them. Image recognition, safety classification, multilingual speech, medical audio, and agents that take a useful job off your team’s plate. Principal AI Engineer · Gold Coast, Australia Explore specialist models: https://invenire.com.au/services/#specialist-models Scope and pricing: https://invenire.com.au/services/#engagement ### Models trained to recognise what matters. I’ve trained image, text, classification, and audio models. I like the whole loop: finding the examples that matter, training something useful, figuring out where it fails, and getting it into people’s hands. - [Image recognition](https://invenire.com.au/services/image-recognition/): Models for the details your team cares about: image categories, subtle defects, and visual conditions that deserve a closer look. - [Safety classification](https://invenire.com.au/services/safety-classification/): Specialist classifiers for safety and policy review, evaluated on missed risks, false alarms, and the cases that need a person. - [Speech & medical audio](https://invenire.com.au/services/speech-medical-audio/): Multilingual speech-to-text on-device, and medical transcription and understanding—including real-time workflows. - [Legal & R&D review](https://invenire.com.au/services/legal-research-review/): Document classification, research evidence checks, and legal review workflows with findings linked to their sources. ### Where should we start? If you’re weighing a decision, start with a review. If you already know what needs building, I can work alongside your team or take on a small build myself. - [AI Production Review](https://invenire.com.au/services/ai-production-review/): A$1,250. Fixed fee · 75-minute working session. Get a second opinion on one prototype, model, or architecture decision, with a written recommendation on what I’d do next. - [30-Day Build Advisory](https://invenire.com.au/services/build-advisory/): A$5,500. Fixed fee · 30 days. I’ll pair with your team on one milestone, with three working sessions and feedback between calls. - [First Agent Pilot](https://invenire.com.au/services/agent-pilot/): From A$12,500. Quoted after scoping. I’ll build one recurring job for your team, with a useful report, a clear owner, and a way to pause it. - [Custom Models & Local AI](https://invenire.com.au/services/model-development/): A$5,500. Fixed fee · model feasibility sprint. Image recognition, safety classification, multilingual on-device speech, and medical audio. Start with one task and a measured comparison. ### I like to start with a real example. Show me an image, a recording, a model result, or a task someone repeats every week. We’ll work out what is worth testing and what your team needs to use the result. ### AI strategy and architecture. Choose a practical first build, compare approaches, and make the important decisions before committing the team. ### Production AI systems. Connect models, retrieval, and agents to your tools, with clear permissions, human review, and a way to see failed runs. ### Evaluation and instrumentation. Turn recurring failures into repeatable checks. Track quality, latency, and cost when a model, prompt, or dataset changes. ### AI-native engineering practices. Pair on real work with coding agents and establish a shared approach to context, review, testing, and accepting changes. ### The work behind the offer. Read the model-training, annotation, research infrastructure, and shipped-product case studies: https://invenire.com.au/work/ ### Bring the awkward examples. Tell me what you’re trying to build and where you’re stuck. A few lines will do. We’ll agree how to share any sensitive material before you send it. Discuss your project: https://invenire.com.au/contact/ Book a quick call: https://invenire.com.au/contact/#book --- ## AI Production Review URL: https://invenire.com.au/services/ai-production-review/ ### Would you ship it? Your AI works in the demo. Now you have to decide whether to put it in front of customers. I’ll look at one prototype, model, or design before we meet. Then we’ll spend 75 minutes on what might break and what to do about it. **A$1,250** · Fixed fee · 75-minute working session ### Your written recommendation - **Decision**. What I’d ship today - **Priorities**. What I’d fix first - **Checks**. How we’d know the fixes worked ### Bring the thing you’re unsure about. An architecture sketch, a short walkthrough, or a few examples that went sideways are plenty to start. Show me where you’re hesitating. We’ll work through it together. - **An image model before rollout**. Inspect a few misses and false alarms, then decide which evaluation gaps matter before the model reaches users. - **A speech pipeline that looks good on paper**. Review the proposed hardware, language coverage, latency measurements, and how draft transcripts reach a person. - **An agent handling sensitive documents**. Review its data access, source references, failure behaviour, and which actions need someone’s approval. ### What’s included. 1. **Send me the short version**. Tell me what you’re deciding and what you’ve tried. I’ll let you know what would help me prepare. 2. **Look at what breaks**. We’ll spend 75 minutes on the design, the failure cases, and what you need to learn before you ship. 3. **Get my recommendation**. I’ll send a one-page note with what I’d do next, in what order, and how I’d check it. ### Analytics Agent. I built Analytics Agent for Shopify merchants. It turns store and analytics data into a brief that explains what changed and what to look at next. [See Analytics Agent](https://analytics-agent.app/) ### A couple of things you might be wondering. - **Can I bring a model or a fine-tuning question?** Yes. Bring one task, what you’re using now, and a few misses. I’ll help you decide what to try and what would count as better. - **What happens when I request a review?** The link opens an email with a few prompts. I’ll reply to confirm whether the review fits, the fee, and when we can meet. ### Send me the short version. Tell me what you’re building and where you’re stuck. A few lines will do. Request a production review: hello@invenire.com.au → Book a discovery call: https://invenire.com.au/contact/#book --- ## 30-Day Build Advisory URL: https://invenire.com.au/services/build-advisory/ ### Let’s get one thing shipped. Bring me into your build for 30 days. Your team leads the work, and I’ll pair with you in three working sessions, review the tricky bits between calls, and help you reach one milestone. **A$5,500** · Fixed fee · 30 days ### One agreed milestone - **Scope**. One feature, model, or workflow - **Working rhythm**. Three sessions and a shared review channel - **Wrap-up**. What’s ready and what’s next ### You’ve started building. Let’s get unstuck. This works best when someone on your team owns the build. Bring a feature you want to get into users’ hands, a technical decision you keep circling, or a model whose behaviour you need to pin down. - **Your first specialist-model release**. Help your team define representative tests, review model changes, and agree when a release is ready. - **A speech workflow with a real deadline**. Pair on the path from incoming audio to a reviewable transcript, with timing and remaining errors visible. - **Coding agents your team can work with**. Pair on a real change and agree how you’ll review, test, and accept the work. ### What’s included. 1. **Pick one milestone**. We’ll define what we want to see working by the end of the month and what we can leave for later. 2. **Pair and review**. We’ll meet three times. Between calls, I’ll review the work we agreed on and aim to reply within two Australian business days. 3. **Check what’s ready**. We’ll check the work against the milestone and write down what to do next. ### Plant. I built Plant to run my own engineering work with agents. I wanted a way to see what they’d actually finished without digging through a chat log. [Read the Plant case study](https://invenire.com.au/work/plant/) ### A couple of things you might be wondering. - **Who writes the code?** Your team owns the code and delivery. I pair with you, review key changes, and help make technical decisions. If you need me to own implementation, we can scope a build. - **How much work can we bring?** We agree on one milestone and the work I’ll review before we start. Larger changes are scoped separately. ### Send me the short version. Tell me what you’re building and where you’re stuck. A few lines will do. Discuss build advisory: hello@invenire.com.au → Book a discovery call: https://invenire.com.au/contact/#book --- ## First Agent Pilot URL: https://invenire.com.au/services/agent-pilot/ ### Give an agent a job you’d hand over. Think of the thing someone has to check every Monday. I’ll build one agent workflow in your tools to do that job, report what it found, and ask for help when it needs a person. **From A$12,500** · Quoted after scoping ### Example: a research review pack - **Reads**. Authorised project records and an agreed checklist - **Reports**. Supporting evidence, source links, and missing material - **Asks for help**. When records conflict or a reviewer needs to decide ### What does someone have to gather and check? Start with a recurring job your team already understands. A research team might need to assemble evidence before a model release. An operations lead might need a weekly brief. We’ll define one useful output and who reviews it. - **Research evidence packs**. Gather experiment records, model versions, and required supporting material. Link findings to sources and flag what is missing. - **Document review preparation**. Organise authorised documents against a checklist and send potential gaps to a named reviewer. The responsible expert makes the final judgement. - **Operational briefs**. Check the data your team already uses and produce a short report on what changed, with links back to the evidence. ### What’s included. 1. **Describe the job**. We’ll agree on when it runs, what it reads, what a good result looks like, and who gets the report. 2. **Try it on real examples**. I’ll start with read-only access where practical. We’ll check a normal run, a missing input, and a case that needs a person. 3. **Hand your team the controls**. You’ll get the workflow, a way to spot a failed run, and instructions to change the schedule or pause it. ### Analytics Agent. With Analytics Agent, I built a Shopify app that runs checks on a schedule and gives merchants a short, plain-English brief. The brief tells them where to look and why. [See Analytics Agent](https://analytics-agent.app/) ### A couple of things you might be wondering. - **Can the agent make changes?** We agree on permissions during scoping. A first pilot usually reads and reports. If the job needs to change something, we agree who can approve the action and how it’s recorded. - **What affects the quote?** The number and complexity of integrations, the quality of the source data, and whether the workflow needs approvals or writes. We confirm scope, acceptance criteria, and recurring costs up front. ### Send me the short version. Tell me what you’re building and where you’re stuck. A few lines will do. Discuss an agent pilot: hello@invenire.com.au → Book a discovery call: https://invenire.com.au/contact/#book --- ## Custom Models & Local AI URL: https://invenire.com.au/services/model-development/ ### A model with a particular set of skills. I’ve trained image, text, classification, and audio models. I help teams recognise the details that matter, understand specialist speech, and make useful decisions from their data—with a model we can test and a system you can run. **A$5,500** · Fixed fee · model feasibility sprint ### Models trained to recognise what matters. I’ve trained image, text, classification, and audio models. I like the whole loop: finding the examples that matter, training something useful, figuring out where it fails, and getting it into people’s hands. - [Image recognition](https://invenire.com.au/services/image-recognition/): Models for the details your team cares about: image categories, subtle defects, and visual conditions that deserve a closer look. - [Safety classification](https://invenire.com.au/services/safety-classification/): Specialist classifiers for safety and policy review, evaluated on missed risks, false alarms, and the cases that need a person. - [Speech & medical audio](https://invenire.com.au/services/speech-medical-audio/): Multilingual speech-to-text on-device, and medical transcription and understanding—including real-time workflows. - [Legal & R&D review](https://invenire.com.au/services/legal-research-review/): Document classification, research evidence checks, and legal review workflows with findings linked to their sources. ### Bring the awkward examples. A subtle defect. A safety label that changes with context. A medical term the transcript keeps getting wrong. We’ll use representative examples to define the task and decide what is worth improving. - **Start with the data**. Check whether the examples represent the job, the labels mean the same thing to your reviewers, and the test set stays separate from training. - **Train where it helps**. Compare an existing approach with one candidate. That could be a specialist classifier, a fine-tuned model, or a simpler approach that already does the job. - **Run it where you need it**. Evaluate local, on-device, or private-cloud deployment against your hardware, latency, access, and operating requirements. ### What’s included. 1. **Look at the examples**. I’ll review a sample with you, agree how to score the results, and set aside examples for testing. 2. **Run a fair comparison**. I’ll measure your current approach and one alternative using the same evaluation criteria. 3. **Decide whether it’s worth building**. You’ll get the scores, the failure cases that matter, and my recommendation on what to try next and what it might cost. ### Hands-on model development. **A$18,000–30,000**. For one task with usable data, I’ll train or fine-tune a model and test it on examples it hasn’t seen. Your team gets the results, the training code, and the model or deployment package. Indicative range for one bounded model task with usable data. Substantial labelling, compute, device integration, and production or clinical-system integration are quoted separately. Discuss a model build: hello@invenire.com.au ### Autonomous model-training agent. I’ve built a system for running repeatable model experiments and reviewing what changed from one run to the next. The case study covers the engineering behind it. [Read the case study](https://invenire.com.au/work/autonomous-training/) ### A couple of things you might be wondering. - **Do I already need labelled data?** Bring what you have. The sprint can establish whether your examples are usable and what labelling is needed. I scope a larger labelling effort separately. - **Does local AI always need a custom model?** No. Where a model runs and whether it needs training are separate decisions. We can test an existing model locally before deciding whether adaptation is worth it. - **What if the alternative performs worse?** You get the comparison and the reasons it fell short. We can use that result to keep the current approach, improve the data, or choose a better candidate to test. ### Send me the short version. Tell me what you’re building and where you’re stuck. A few lines will do. Discuss a model feasibility sprint: hello@invenire.com.au → Book a discovery call: https://invenire.com.au/contact/#book --- ## Image recognition URL: https://invenire.com.au/services/image-recognition/ ### Teach it what to look for. A defect, a species, a detail in an inspection photo. I train and adapt image models around the things your team needs to recognise, then test them on the awkward examples as well as the obvious ones. Illustration: A closer look at the detail that matters. ### Your categories. Your real-world conditions. I’ve trained image and classifier models, and built the tooling around data review and model experiments. I enjoy the whole loop: making the labels useful, trying an approach, finding its weak spots, and getting it running where it needs to work. - **Visual inspection**. Classify defects or conditions in product, equipment, or site photos. Send unclear images back to a reviewer with the original image attached. - **Specialist image collections**. Organise research imagery, camera-trap photos, or technical archives using categories defined by the people who know the domain. - **Visual safety checks**. Flag potentially unsafe conditions for review. Agree which things the images can actually establish, and which need a person on site. ### A scratch, a shadow, or something new? An inspection team has years of photos, but small defects are easy to confuse with reflections. A useful first build separates known conditions and makes unfamiliar examples easy to review. Example engagement. Illustrative workflow, not a client result or live model demonstration. - **Input**. Inspection photos and examples reviewed by your specialists. - **Model task**. Suggest a category; flag uncertain or unfamiliar images. - **Human decision**. The reviewer checks the original image and records the final label. ### What we’d measure. - **Misses and false alarms**. Measure each important class separately, including defects that are rare but costly to miss. - **New conditions**. Test on unseen sites, cameras, lighting, and capture periods—not near-duplicates of training images. - **The actual device**. Check speed, memory, and offline behaviour on the hardware you intend to use. The inspection criteria come from your domain experts. We test the model against them and define when it must defer to a person. ### The data work matters as much as the model. My image annotation and QA work is about making human review useful for training. The anonymised case study describes that part of the engineering. [Read the annotation case study](https://invenire.com.au/work/annotation-platform/) ### Start with one useful test. Bring one task, a few representative examples, and the constraints that matter. We’ll agree what to compare before starting. Model feasibility sprint: A$5,500. One task, a baseline, and one alternative. You get the comparison, the important failure cases, and a recommendation on what to build or keep. [Scope and build pricing](https://invenire.com.au/services/model-development/#scope). Need an initial second opinion? The A$1,250 production review is a smaller place to start. Larger builds and production integration are scoped separately. ### A few useful questions. - **Can it run locally?** Potentially. We choose a candidate around the task and target hardware, then measure it there. Local deployment and model training are separate decisions. - **What if our labels are inconsistent?** That is useful to discover early. I’ll help identify disagreement, clarify the categories with your specialists, and scope any additional labelling before a larger training run. ### Bring the awkward examples. Tell me about the task and where the current approach falls short. A description is plenty for the first conversation; we’ll agree how to share any sensitive material. Discuss your project: hello@invenire.com.au Book a quick call: https://invenire.com.au/contact/#book --- ## Safety classification URL: https://invenire.com.au/services/safety-classification/ ### Context changes the answer. The same words can mean different things in a threat, a quotation, or a conversation about safety. I build and evaluate classifiers around those distinctions, with clear categories and room for uncertainty. Illustration: The signal makes more sense in context. ### A useful decision starts with a useful definition. Bring your policy, examples your reviewers agree on, and the ones they debate. My classifier and training experience helps turn those judgements into a task we can test, including the trade-off between catching a concern and raising an unnecessary alarm. - **Context-sensitive content review**. Classify messages or passages against a defined safety policy, accounting for quotations, discussion, and surrounding context. - **Policy-specific categories**. Adapt a classifier to your review categories and escalation paths, rather than forcing your workflow into a generic set of labels. - **A manageable review queue**. Prioritise material for specialists and keep uncertain cases visible, with the relevant context attached. ### Similar words. Different situations. A product team wants to flag concerning messages without treating a safety discussion as a direct threat. The first task is a reviewer-labelled set of contrasting examples, with enough context to judge each one. Example engagement. Illustrative workflow, not a client result or live model demonstration. - **Input**. A message, relevant surrounding context, and the applicable policy version. - **Model task**. Suggest a review category or defer when the case is unclear. - **Human decision**. A trained reviewer checks escalated cases and decides what action is appropriate. ### What we’d measure. - **Missed concerns**. Review the cases the system fails to flag, broken down by category and consequence. - **Unnecessary alerts**. Measure false alarms and the resulting reviewer workload, including benign discussions of sensitive topics. - **Changes in context**. Test quotations, slang, language variation, and inputs unlike the training set. Recheck when the policy or model changes. A classifier supports a defined review process. It does not establish someone’s intent or make a complete safety assessment on its own. ### Start with the examples that disagree. I’ve worked on model training and data-quality workflows. That experience is useful here: clarify what the label means, preserve reviewer decisions, and test changes against a consistent set of examples. [Read about my model-training work](https://invenire.com.au/work/autonomous-training/) ### Start with one useful test. Bring one task, a few representative examples, and the constraints that matter. We’ll agree what to compare before starting. Model feasibility sprint: A$5,500. One task, a baseline, and one alternative. You get the comparison, the important failure cases, and a recommendation on what to build or keep. [Scope and build pricing](https://invenire.com.au/services/model-development/#scope). Need an initial second opinion? The A$1,250 production review is a smaller place to start. Larger builds and production integration are scoped separately. ### A few useful questions. - **Can we tune the balance between misses and alerts?** Yes. We compare thresholds on held-out examples and show the trade-offs, including how much material would reach reviewers. Model confidence alone is not proof that a decision is correct. - **Can the model make enforcement decisions?** We define that boundary during scoping. A first pilot can classify and route material for human review, with actions and access handled by the surrounding software. ### Bring the awkward examples. Tell me about the task and where the current approach falls short. A description is plenty for the first conversation; we’ll agree how to share any sensitive material. Discuss your project: hello@invenire.com.au Book a quick call: https://invenire.com.au/contact/#book --- ## Speech & medical audio URL: https://invenire.com.au/services/speech-medical-audio/ ### Every word. In context. I work on multilingual speech-to-text, on-device models, and medical audio—including real-time transcription and understanding. The interesting details are the terminology, the language switches, and what happens when someone corrects themselves mid-sentence. Illustration: From speech to meaning, on the hardware that matters. ### Built around the conversation. I’ve trained audio and voice models. I can help adapt and evaluate a speech pipeline for your languages, vocabulary, devices, and workflow—from a local transcript to structured information a clinician can check. - **Multilingual, on-device transcription**. Speech-to-text for selected languages and language switching, with latency, memory, and offline operation tested on your target device. - **Medical dictation**. Transcribe specialist terminology, abbreviations, medication names, and corrections into a draft for clinician review. - **Real-time medical understanding**. Turn a live conversation into a transcript and draft structured note, with extracted details linked to the relevant audio for checking. ### The correction matters as much as the words. A clinician dictates a note, corrects a detail, and switches language for a name or phrase. The workflow needs to preserve the intended meaning while making it easy to check what was actually said. Example engagement. Illustrative workflow, not a client result or live model demonstration. - **Input**. Authorised audio, the languages in scope, and the agreed note structure. - **Model task**. Produce a transcript and draft structured details, retaining links to the audio. - **Human decision**. The clinician checks and edits the draft before it becomes part of the record. ### What we’d measure. - **Meaning, not only word count**. Test negation, medication names, dosages, speaker attribution, and self-corrections alongside transcription errors. - **Keeping up**. Measure time to the first text, revisions while someone speaks, and time to the final reviewable transcript or note. - **Real listening conditions**. Evaluate the languages and accents in scope, background noise, overlapping speakers, and sustained performance on the target device. We scope transcription, understanding, and note generation separately. Draft clinical content requires clinician review; processing location, recording consent, retention, and access are agreed with your team. ### Let’s test the audio you actually hear. My experience includes training audio and voice models. For a new engagement, I start with representative recordings and the hardware constraints, then compare approaches before committing to a larger build. [See the model feasibility sprint](https://invenire.com.au/services/model-development/#scope) ### Start with one useful test. Bring one task, a few representative examples, and the constraints that matter. We’ll agree what to compare before starting. Model feasibility sprint: A$5,500. One task, a baseline, and one alternative. You get the comparison, the important failure cases, and a recommendation on what to build or keep. [Scope and build pricing](https://invenire.com.au/services/model-development/#scope). Need an initial second opinion? The A$1,250 production review is a smaller place to start. Larger builds and production integration are scoped separately. ### A few useful questions. - **Will everything run on-device?** We test that rather than assume it. Speech recognition, speaker handling, and note generation may have different hardware needs. We agree which parts must stay on-device and measure the complete proposed pipeline. - **Which languages can we support?** We select the languages, accents, and language-switching patterns during scoping, based on the available models and representative evaluation audio. Support for a language does not establish medical accuracy in that language. - **Does real-time mean the text never changes?** No. Streaming transcripts can revise earlier words as context arrives. We measure those revisions and the finalisation delay, so the interface distinguishes provisional text from the version ready for review. ### Bring the awkward examples. Tell me about the task and where the current approach falls short. A description is plenty for the first conversation; we’ll agree how to share any sensitive material. Discuss your project: hello@invenire.com.au Book a quick call: https://invenire.com.au/contact/#book --- ## Legal & R&D review URL: https://invenire.com.au/services/legal-research-review/ ### Keep the evidence in view. I build tools that help specialists find the material worth a closer look: a clause outside the playbook, a missing research record, or a document that needs restricted handling. Each finding should take you back to the evidence. Illustration: A finding is more useful when you can follow it back. ### Your experts define the rules. We make them testable. My background includes research infrastructure, model experiments, and data-quality systems. I bring that engineering to review workflows: clear definitions, reproducible checks, and a record of what a person verified. - **Legal playbook checks**. Classify clauses, extract obligations, and flag potential departures from an approved review playbook, with the original passage attached. - **R&D evidence packs**. Check research records against an agreed checklist: dataset provenance, approvals, experiment records, model versions, and missing supporting material. - **Sensitive document handling**. Suggest document categories and flag potentially sensitive passages for review before material is shared or released. ### A review pack you can follow back to the source. An R&D team is preparing a model release. The results look promising, but the reviewer also needs to see which dataset, approvals, code, and model version support the result. An agent can gather the material while a specialist model classifies or extracts what the checks need. Example engagement. Illustrative workflow, not a client result or live model demonstration. - **Input**. The agreed checklist and authorised access to the project’s records. - **System task**. Assemble source-linked evidence; flag missing records or potential inconsistencies. - **Human decision**. The responsible expert reviews the evidence and decides whether the requirements have been met. ### What we’d measure. - **Traceable findings**. Check that each finding points to the correct passage or record, and the exact source version. - **Missing evidence**. Try incomplete packs, conflicting records, and unfamiliar documents. A missing source must remain visibly missing. - **Review boundaries**. Test permissions, separation between matters or projects, and which decisions require an authorised reviewer. These tools support legal and R&D compliance review. They do not certify compliance or replace legal, clinical, or governance judgement. Local hosting is a deployment choice, not a compliance guarantee. ### Research infrastructure is part of my background. I was Technical Lead and Solutions Architect on EcoCommons. Building for researchers taught me to treat provenance, reproducibility, and usable workflows as part of the product. [Read the EcoCommons case study](https://invenire.com.au/work/ecocommons/) ### Start with one useful test. Bring one task, a few representative examples, and the constraints that matter. We’ll agree what to compare before starting. Model feasibility sprint: A$5,500. One task, a baseline, and one alternative. You get the comparison, the important failure cases, and a recommendation on what to build or keep. [Scope and build pricing](https://invenire.com.au/services/model-development/#scope). Need an initial second opinion? The A$1,250 production review is a smaller place to start. Larger builds and production integration are scoped separately. ### A few useful questions. - **Does this always need fine-tuning?** No. Retrieval, rules, and an existing model may be enough. Training earns its place when a comparison shows a useful improvement on a recurring task. - **Can it work in our own environment?** We can scope local or private deployment around your access, retention, and infrastructure requirements. Your security and governance team helps define the requirements; the build is evaluated against them. ### Bring the awkward examples. Tell me about the task and where the current approach falls short. A description is plenty for the first conversation; we’ll agree how to share any sensitive material. Discuss your project: hello@invenire.com.au Book a quick call: https://invenire.com.au/contact/#book --- ## Selected work URL: https://invenire.com.au/work/ ### Hero **Systems that made it past the demo.** Selected work across AI agents, analytics, autonomous engineering, national research infrastructure, climate platforms, model-training loops, and audit-grade data workflows. The common thread is not the domain. It is the level of responsibility: systems that need to produce useful outputs repeatedly, under constraints, with enough structure to be trusted by real users. ### Case study — Analytics Agent **Tag:** SHOPIFY · LIVE · 2025– · https://analytics-agent.app A production Shopify analytics product that turns GA4, Search Console, and store data into daily merchant decisions. Six core analysis agents are coordinated by a router to detect anomalies, surface opportunities, audit GA4 configuration, and produce a plain-English Mission Brief. The product demonstrates Invenire's core operating pattern: bounded agents, structured outputs, recurring evaluation, operational telemetry, and commercial UX wrapped around the model layer. ### Case study — Plant — autonomous engineering platform **Tag:** SOLO-BUILT · SELF-HOSTING · 2025– **Link:** https://plnt.dev **URL:** https://invenire.com.au/work/plant/ An autonomous engineering platform exploring how one senior operator can coordinate AI agents, externalised state, objective verification, and multi-service software delivery. Seven services across five languages, designed around agent-first workflows and completion checks rather than ad hoc prompting. Plant is the R&D layer behind Invenire's AI-native engineering practice: how to use agents aggressively without losing architecture, reviewability, or control. The decisions that make it work: externalised state beats context accumulation (conversation is deleted between iterations); objective completion verification, not the model's word; tooling built for agents, not adapted from human tools; Code Mode (a typed TypeScript API generated from MCP schemas); a system that watches itself; and a system that builds itself, every repo developed using the loop. What it is: a real, self-hosting system doing real work. What it isn't: a commercial product — test coverage is deliberately uneven and several conveniences suit a solo operator. **Signal:** services=7 · languages=5 (PHP, Go, Rust, Swift, TypeScript) · operator=1 · self-hosting=every repo built by the loop it runs **Scale:** Spark ~7,000+ test cases · Ralph ~24k LOC, 470+ tests · Foreman ~58k LOC, 1,000+ test functions · Chief ~24k LOC · Telemetry 5-crate Rust core **Tech:** Laravel, React, Go, Rust, Swift, Docker, AWS spot, Pulumi, MCP, Claude Code, PostgreSQL/pgvector, SQLite/Tantivy ### Case study — EcoCommons — national ecological modelling **Tag:** TECHNICAL LEAD & SOLUTIONS ARCHITECT · GRIFFITH UNIVERSITY · 2020–2023 **Link:** https://ardc.edu.au/project/ecocommons **URL:** https://invenire.com.au/work/ecocommons/ National-scale ecological and environmental modelling infrastructure. As Technical Lead and Solutions Architect at Griffith University, Arve helped deliver a reproducible modelling platform across nine partner institutions and a multi-year funding environment. This work matters because production AI has the same hard requirements as serious research infrastructure: provenance, reproducibility, permissions, reliability, and users who need to trust the output. The hard part was making national-scale, reproducible science work across nine partner organisations, four scientific disciplines, and a decade of incompatible legacy systems (BCCVL, ecocloud, CSDM re-engineered into one platform), wiring in national and global data providers (ALA, TERN, GBIF, OBIS, IMOS) and giving researchers a reproducible-by-design cloud sandbox. **Signal:** role=Technical Lead & Solutions Architect · scope=national, 9 partner institutions · program=3-year, 2020–2023 · domains=biodiversity, biosecurity, hydrology, agriculture **Tech:** distributed cloud + HPC, microservice architecture, species-distribution & climate models, FAIR data pipelines, JupyterLab / virtual-desktop sandboxes, reproducible-by-design environments ### Case study — Autonomous model-training agent **Tag:** CONFIDENTIAL CLIENT · APPLIED AI · 2025–2026 · Reference available under NDA **URL:** https://invenire.com.au/work/autonomous-training/ A confidential agentic system that runs repeated ML experimentation loops: hypothesis, code, training run, scoring, analysis, and next experiment. The hard work is not making the agent act. It is preventing reward hacking, false progress, wasted compute, and unreviewable changes. The engineering challenge was designing the loop so progress could be trusted. The engineering is all in the guardrails: an evaluation harness the agent cannot edit; git as the experiment ledger; a deployability rule rather than a bare accuracy score; and a loop built to survive its own length by keeping working memory in files. **Signal:** human ML engineers in loop=none · compute=commodity spot GPUs · reproducibility=git-ledgered · search=architecture chosen per problem **Tech:** Python, PyTorch, timm, agentic coding agent (Claude Code), spot-GPU infrastructure, object storage, git-as-ledger ### Case study — Image annotation and QA platform **Tag:** CONFIDENTIAL CLIENT · PRODUCTION · 2025–2026 · Reference available under NDA **URL:** https://invenire.com.au/work/annotation-platform/ A confidential platform for turning inconsistent human review into audit-grade, model-ready datasets. The system combines weighted consensus, disagreement surfacing, provenance, and QA workflows so the resulting data can be trusted downstream. The value is not automation for its own sake. It is higher-confidence data production under real operational constraints. **Signal:** build=solo, end to end · role=design, architecture, implementation · trust model=weighted consensus, disagreement surfaced, full provenance · output=human-verified, model-ready datasets **Tech:** Next.js, React, TypeScript, PostgreSQL, cloud data pipeline, SSO, object storage ### Earlier shipped work Before the frontier work: document-intelligence and retrieval systems for clients with strict compliance and audit requirements, and twenty years of commerce platforms, payment integrations and web infrastructure. ### CTA **Want something like this built?** Thirty minutes. → Discuss an engagement: https://invenire.com.au/contact/ --- ## About URL: https://invenire.com.au/about/ ### Hero **AI engineering for the part after the demo.** Tag: principal_ai_engineer · founder I'm Arve, founder of Invenire. I help teams design, evaluate, and ship AI systems that need to work in production — agents, retrieval, vision, analytics, automation, and AI-native engineering workflows. ### Founder bio — twenty years, three eras of stack I have spent 20 years shipping production software, from national-scale research infrastructure to modern agentic AI systems. Before founding Invenire, I was Technical Lead and Solutions Architect on EcoCommons, Australia's national platform for ecological and environmental modelling, and Technical Lead on the Pacific iCLIM project with Griffith University and SPREP. I’ve trained image, text, classification, and audio models. My work includes multilingual speech-to-text on-device, medical transcription and understanding, and the evaluation and software around specialist models. I also build agents and help teams get their AI systems into production. I stay close to frontier models and tools because the ground is moving quickly. But I am not interested in novelty for its own sake. The test is whether a system becomes more useful, reliable, faster, cheaper, safer, or easier to operate. ### Why clients bring me in Clients usually bring me in when AI has become important enough that experiments are no longer enough. They need someone who can speak strategy with leadership, architecture with engineers, and implementation with the codebase open. - **The prototype works, but nobody trusts it** — I help teams add evals, telemetry, review loops, and architecture around fragile AI prototypes. - **The opportunity is clear, but the path is not** — I help teams decide what to build, what to buy, what to ignore, and where AI can create real leverage. - **The team is moving fast, but without a system** — I help engineering teams adopt AI-native workflows without sacrificing quality, architecture, or review. - **The work is too important for generic advice** — I work best where the problem needs both senior engineering judgement and current AI capability. ### Career arc — The same bar, three eras of stack. - **The frontier work (2024–).** Agents, evals, retrieval and vision in production: Analytics Agent for Shopify merchants, the Plant autonomous engineering platform, and two confidential applied-AI client systems — an autonomous model-training agent and an enterprise annotation and QA platform. The systems are on the work page; the judgment behind them is what I sell. - **National-scale infrastructure (2020–2023).** Technical Lead and Solutions Architect for EcoCommons (https://ardc.edu.au/project/ecocommons), Australia's national platform for ecological and environmental modelling — a three-year program across nine partner institutions at Griffith University. Designed the platform and led the engineering team and partners through to launch. - **Research infrastructure (before 2020).** Technical Lead on the Pacific iCLIM project — a Griffith University and SPREP (https://www.sprep.org) collaboration funded by Australia's Department of Foreign Affairs and Trade. Led delivery of the climate-change decision-support tools behind SPREP's Pacific Climate Change Portal, including the Adaptation Project Tool and the Climate Finance Navigator, and ran the training that put them in the hands of government officials across the Pacific. The stack changed — the bar didn't. ### Operating principles — Four non-negotiables. - **currency** — The AI frontier changes quickly. I track the models, tools, patterns, and failure modes closely, then apply the parts that hold up under production constraints. - **evidence** — AI systems need measurement. I prefer evals, traces, golden datasets, regression gates, cost curves, and failure taxonomies over opinions about whether a model "seems better." - **reliability** — Non-determinism is not an excuse. Production AI needs bounded responsibilities, structured outputs, observability, fallback paths, human review where appropriate, and tested recovery. - **focus** — The goal is not to use the most AI. The goal is to ship the smallest system that creates leverage and can be trusted. ### Stack (`$ arve --stack --json`) ``` { "languages": ["TypeScript", "Node.js", "Python", "PHP", "SQL", "Rust", "Go", "Bash"], "ai": ["LLM orchestration", "multi-agent systems", "retrieval pipelines", "eval frameworks", "vector indexing", "geospatial", "vision"], "infra": ["AWS", "Azure", "GCP", "Docker", "Kubernetes", "Terraform", "GH Actions"], "databases": ["Postgres", "MySQL", "MongoDB", "DuckDB", "Snowflake", "Vector stores"], "specialties": ["Agentic AI", "Enterprise security & compliance", "Client-facing technical delivery"] } ``` ### Formal training - **2006** — Master of Information & Communication Technology — Australia - **2005** — Bachelor of Information Technology — Australia - **2001** — IT Studies — Norway ### References Available on request. EcoCommons and iCLIM relationships are real and reachable. ### Contact - **studio_inbox / Email** — hello@invenire.com.au - **network / LinkedIn** — linkedin.com/in/arvesolland - Founder-led. You work directly with Arve. --- ## Contact URL: https://invenire.com.au/contact/ ### Hero **Discuss a production AI engagement.** Pill: currently open · limited capacity Tell me what you’re building and where you could use a second pair of eyes. I work directly with founders and teams on production reviews, agents, and model development. CTAs: Pick a time (Cal.com inline embed on page) · Email instead (hello@invenire.com.au) ### Discovery call Pick a time. Thirty minutes, no slides. We'll talk through what you're building, where it's stuck, and whether Invenire is the right fit. Send the short version by email first if that's easier — the next step is a focused architecture call. The booking calendar is the inline Cal.com embed on the page. Prefer email? hello@invenire.com.au. ### Channels - **email** (hello@invenire.com.au) — Primary channel. Best for project enquiries, scoping calls and anything you'd like a written trail of. - **linkedin** (linkedin.com/in/arvesolland) — Track record, recommendations and the longer history. - Founder-led. You work directly with Arve. ### What to include in your first message To make the first conversation useful, include: - The system, workflow, or opportunity you are working on - What already exists: idea, prototype, production system, team, data, or codebase - What is currently blocking progress - What changes if this works - Your timeframe - Your rough budget range or commercial constraints - Whether you need strategy, implementation, review, or team enablement ### Fit **Good fit:** - You have a serious AI opportunity and need senior technical judgement. - You are moving from prototype to production. - You need evals, telemetry, architecture, or review before scaling. - You want a small senior studio, not a large consulting team. - You value working software over slide decks. **Not a fit:** - You want a generic chatbot with no clear business owner. - You need low-cost development capacity. - You are looking for AI theatre rather than operational change. - You are not ready to give access to the people, systems, or context required to do the work properly. ### Scope and investment See the [consulting engagements and prices](https://invenire.com.au/services/#engagement). For custom builds, I’ll confirm scope and fees once I understand the work. Invenire is a small, founder-led studio and only takes on a limited number of engagements at a time. The best fit is work where senior engineering judgement matters. ### Quick answers (FAQ) - **q / capacity — Are you taking new engagements?** Yes — limited capacity. Typically 1–2 clients at a time. - **q / location — Do you work remote or onsite?** Both. Remote default; onsite trips possible globally. - **q / minimum — What's the minimum engagement?** An AI Production Review is a 75-minute session with advance review and a written recommendation. Larger engagements are scoped to the work. - **q / nda — Do you sign NDAs?** Yes, for serious conversations. Send yours or use ours. ### Response time **Replies within 48 hours.** Australian business days. If it's urgent, say so in the subject line and we'll prioritise. → hello@invenire.com.au --- © 2026 Invenire · Gold Coast, AU