Vindex.PI
70 years of IP records that lived only in scanned PDFs, now a structured, searchable database with a layered LLM verification pipeline.
WORK
LLM and data systems I built and shipped to production — where the information was broken and someone real uses it today.
Scattered data, dead PDFs, incompatible sources — turned into systems someone real uses. Live.
70 years of IP records that lived only in scanned PDFs, now a structured, searchable database with a layered LLM verification pipeline.
104+ environmental variables scattered across public sources, joined into a per-municipality territorial-risk score that did not exist before.
The coffee supply chain runs in silos. Turns pollination and traceability data into verifiable biodiversity credits.
Parallel rate, BCV, basket and inflation scattered across sources, unified into an API with a layer that quantifies FX pressure.
The engineering behind the two flagship systems — architecture, model choices, the trade-offs.
A multi-model LLM pipeline that verifies 1M+ IP records against scanned bulletins, built around trust: 92% match-rate on multi-column OCR, automated quality gates. In production, paying client.
A multi-source geospatial ETL joining climate, disease, land-use and census data across 645 municipalities × 104 variables, published FAIR with a DOI and validated in a state-government workshop.
Remote hire or a one-off project. Tell me what you need to build.