Launching Asotele: open economic intelligence for Nigeria
Today we're publicly launching the Asotele project — an open-source AI system for Nigerian and African emerging-market economic forecasting.
For the past six months, Asotele has been operating quietly on homelab infrastructure in Lagos. The current state of the system:
- 17 data scrapers running daily at 04:30 UTC — covering oil, official and parallel FX, CBN monetary policy, NGX equities across 9 sectors (34,349 historical data points back to 1996), food and fuel prices, cement, employment, telecom subscribers, financial news sentiment via three NLP models, Google Trends consumer signals, World Bank macro indicators, and cross-border USDT flows - Three econometric forecast engines: ARIMA, Markov regime-switching with GARCH(1,1), and VAR + Granger causality (activating later this month at 30 observations per series) - Daily economic briefings generated by Ollama-hosted Qwen 3 14B + Mistral Small 24B + Devstral 24B - A GitHub repository and HuggingFace organization going public alongside this site
A Bloomberg Terminal costs ~$32,000 per year. Nigeria's per-capita GDP is ~$1,100. That ratio — roughly 30× — explains why Nigerian journalists, small businesses, and policy researchers operate without access to the analytical tools any mid-sized American business takes for granted. The deeper truth: no private vendor will build a Bloomberg for Africa, because the unit economics don't work at African price points.
Asotele is built as a public good for that reason. Every component will be released under permissive open licenses. The architecture is explicitly replicable to Ghana, Kenya, Egypt, and other African economies in weeks rather than years.
The next 90 days are about three things:
1. Compute grants for foundation-model fine-tuning. Applications are in flight to Google TPU Research Cloud, HuggingFace ZeroGPU, NVIDIA Inception, MIT Solve + the Patrick J. McGovern AI for Humanity Prize, and LINGUA Africa. The single 3090 in our homelab cannot fine-tune Qwen 3.6 35B-A3B or Gemma 4 26B-A4B without external compute. 2. African-language expansion. Pending the LINGUA Africa decision, we plan to add Hausa, Yoruba, Igbo, and Nigerian Pidgin financial NLP — making Asotele accessible to the ~100 million Nigerians currently excluded from English-only economic discourse. 3. Distribution partnerships. Conversations starting with Nigerian newsrooms, university economics departments, and African central banks.
- Subscribe to project updates below — we'll send no more than 2 emails per month - Star the repo at github.com/apexgrid/asotele (live shortly) - Reach out — francis.oyakhire@gmail.com — if you're a newsroom, university, or organization interested in early integration
Thanks for being here at day one.
— Francis Oyakhire, Apex Grid Technologies, Lagos