The hardest part about building EconLens was not the AI. It was the data.
Getting a modern model like GPT‑4‑class systems to produce good analysis is relatively straightforward if you give it clean context. The challenge is building a reliable pipeline that queries five different government APIs with five different data formats and normalizes everything into a coherent context window in under three seconds.
FRED returns JSON. The ECB uses a nested SDMX format. Eurostat has its own JSON structure. OECD changed its API format last year and broke half of the internet’s OECD integrations. World Bank is the most stable but its data often lags by 6–12 months.
Each source has different update frequencies, missing‑data conventions, and error codes. Making all of that invisible to the user is where roughly 80% of the engineering effort went.
The AI is the flashy part. The data pipeline is what makes the tool actually useful.
econlens.app.