AI Labs Community
Where experiments get run, broken and written up.
Compute is scarce and expensive, datasets for local languages and markets are thin, and most experiments fail. The Labs make failure cheap: shared resources, shared notes, and people who have hit the same wall.
Open to every member. Join now and help shape what it builds next.
What we work on
Four problems define this community’s work right now. Every sprint, review and essay here comes back to one of them.
Weekly model sprints
One focused experiment a week, a new architecture, dataset or deployment problem, with results shared openly.
Local data
Building and cleaning datasets for the languages, markets and conditions mainstream models ignore.
Reproducible work
Notebooks, configs and write-ups that someone else can rerun and get the same answer.
Small and efficient models
Getting useful results from models that run on modest hardware and cheap phones.
Questions we are working through
No one has settled these yet. They come up in every sprint, and members argue them out in essays and reviews.
- How much data do you really need before a model beats a simple rule?
- How do you build a fair dataset for a language with almost no text online?
- What is the smallest model that still solves the problem well enough?
Our impact
- Engineering essays published on TechOps News
- 3
- Systems in AI research we have built and shipped
- 2
Shipped in AI research
- ShambaRemote Sensing · Zambia · 2023
Satellite and drone imagery pipelines that flag crop stress early, delivered to agronomists in-field.
11% yield uplift40k ha monitored - MazaoAgri Intelligence · Tanzania · 2023
Crop-yield forecasting and market-price advisory, delivered as a white-label platform and API.
14k farmers served6 crops modelled
From the field
All Engineering essays · 15 min of reading- Retrieval before fine-tuning: making an LLM answer from your own documentsIf you want a language model to answer from your company's documents, start with retrieval, not training. A worked example, from the first document to the first honest measurement.
- Test your AI feature like you would test a payments flowNobody ships a payments change because it looked fine in a demo. AI features deserve the same discipline. A worked example of building an evaluation suite your team will actually use.
- Build for the network your users actually haveYour office Wi-Fi is not your user's connection. Designing for real phones and paid-for data changes what good engineering looks like.
Your first build
New members pick one of these for their first month. Each is small enough to finish and real enough to show an employer or an investor.
A baseline that is hard to beat
Pick a real dataset, build the simplest strong baseline, and write up exactly why it works.
- Python
- Statistics
- Writing
A small local-language dataset
Collect, clean and document a few thousand labelled examples in a language you speak.
- Data collection
- Labelling
- Documentation
A distilled model
Shrink a large model into a small one that runs on a phone, and measure what you lose.
- PyTorch
- Distillation
- Benchmarking
Words you will hear
New here? These come up in your first week. Plain meanings, no jargon.
- Baseline
- The simplest reasonable approach. Any clever model has to beat it to be worth the cost.
- Overfitting
- A model that memorised its training data and fails on anything new.
- Ablation
- Removing one part of a system to measure how much that part actually matters.
- Distillation
- Training a small model to copy a large one, trading a little accuracy for speed and cost.
Who you will meet
- Engineers who like running experiments
- Students and researchers
- Data scientists
- Anyone curious enough to share a failed result
How the community runs
- Every week
Build sprint
Bring a real problem from your work in the sector. Leave with progress, reviewed by people who have shipped the same thing.
- Every week
Office hours
Drop in with a model that will not converge, a regulator question or a hiring decision. Practitioners answer.
- Every month
Demo day
Members show working systems to the community, mentors and invited investors. Demos, not decks.
The team behind it
Lead the AI Labs Community
We are looking for experienced practitioners to lead this community: host one session a month, review members’ work and shape what it builds next.
Tell us about your workJoin the AI Labs Community
One payment of $15 makes you a member for life: every community, its sprints, reviews and events. No subscription.