Vidyut
Fintech infrastructure for EV financing. I built an in-house LLM/OCR document system that cut vendor cost 70%, and architected the bureau/NBFC integrations that automated most of underwriting.
The problem
Lending runs on documents — KYC, bank statements, vehicle papers — and on data from credit bureaus and NBFC partners. Vidyut's financing flows were paying a vendor for document extraction and stitching integrations by hand. Both were slow and expensive.
What I built
In-house document intelligence
I replaced the third-party extraction vendor with an in-house LLM/OCR document system. It cut vendor cost by 70% and brought processing time down from 2–3 minutes to under 30 seconds per document.
- 70%
- 2–3min → <30s
- in-house
Bureau & NBFC integrations + ops products
I architected the credit-bureau and NBFC integrations and the internal operations products around them. That automated 60% of underwriting, improved turnaround time by 40%, and lifted conversion from 45% to 55%.
- LLM/OCR extraction, in-house
- Credit bureaus + NBFC partners
- Internal underwriting & operations tooling
- Faster, cheaper, higher-converting lending flow
What I took from it
Most of the value in applied AI is plumbing. The model was the easy part; the win came from wiring it into a real workflow — documents in, decisions out — and measuring the thing that mattered: cost, time, and conversion.
Next — Starship