This week
TBA — Speaker TBA
Topic and speaker announced in the community chat. Come with questions — every talk ends in open discussion, and that's usually the best part.
Agents Revisited, pt. 2
Samuel Ahuno — Part two of last week's session — picking up where the first pass left off, and going further into the parts that only show up once an agent is doing real work.
Learn by poking at it
Interactive explainers — every figure measured, every source linked. Built for people without a CS degree.
Foundation models: one model, five ways to use it
One question orders the whole field: do the model's numbers change? 21 models, 7 fields.
explore →The compute gap: what an AI GPU costs from Accra
The same 8×H100 machine costs 1.72× more here than in Virginia. From the providers' own price lists.
explore →The token tax: why AI costs more in Twi and Ga
The same sentence costs 2.55× more in Ga than in English — and Ga is the shorter text.
explore →Temperature 0 was never deterministic
30 identical requests, seed pinned, 26 different answers. Measured on both major APIs.
explore →The judge has a favourite seat
Show an AI grader two equally good answers and it still picks a winner — then picks the other one when you swap the order.
explore →One flyer, two kinds of image
Our 1.4 MB PNG vs the 843 KB PDF whose words never blur — and the 72 DPI secret inside it.
explore →What happens here
Three ways to grow — most people end up doing all three.
Weekly talks from Ghanaian and diaspora researchers, open discussions where questions are welcome, and a free fellowship — no CS degree required, life scientists especially welcome.
build/Real problems, real users. Take on a hard problem with problem owners who'll test your work, compute credits, and diaspora mentors. Ship for 20 real people before you scale.
grow/Shared growth. Mentorship both ways, writing and speaking reps, a venture track for future founders and investors — and a community that opens doors for each other.
Open opportunities
Fellowships, grants, scholarships, hackathons, prizes — one list, sorted by deadline, so nobody misses a door because they didn't know it existed.
| type | mode | opportunity | value | deadline |
|---|---|---|---|---|
| hackathon | hybrid | AMD Developer Hackathon: ACT III | $5,000+ prize pool | 2026-10-18 |
| prize | remote | Climate Risk and Health Prediction Challenge | $1,000 | 2026-10-18 21:59 UTC |
| hackathon | remote | Build, Ship, Shape: Amazon Developer Hackathon | ~$171,000 total — Fire TV and Alexa+ tracks ($25k/$15k/$4k each), Bee and Ring tracks ($12k/$8k), plus $10k in mini-challenges | 2026-10-23 |
| hackathon | remote | Build With AI: Basics — Devpost Learn | $2,500 ($1,250 / $750 / $500) + a Devpost Learn badge for every valid submission | 2026-10-26, 17:00 EDT |
| hackathon | remote | OpenCV AI Competition 2026, powered by AWS | $20,250 total — $5,000 grand / $3,000 / $2,000, plus a $1,000 "Agentic Vision" award and 55 × $150 AWS compute grants | 2026-10-26 |
The problem board
Problems are our curriculum — the most honest way to learn is to build something a real person needs.
African-language speech models
ASR/TTS for Twi, Ewe, Dagbani, Ga — and the pattern generalizes to hundreds of low-resource languages across the continent.
open problem →Voice-note bookkeeping
Spoken Twi market transactions → a ledger a loan officer trusts. No published system does this yet.
open problem →Bio × AI
Our community is life-sciences-heavy. Pair with a genomics or public-health member and build what neither of you could alone.
open problem →This org runs on agents we built
Metrics, application scoring, grant scanning, comms, sponsor briefs, deal memos, the daily AI briefing — seven agents in production, built by our own members. We teach what we practice: even our operations are a learning project you can contribute to.