Live system · Golden Record RAG Pipeline
Fragmented customer data, resolved into one trusted record.
A production-deployed customer-data platform for fintech compliance — entity resolution, survivorship, attribute lineage, vector RAG and GDPR tooling, running end-to-end on 89,198 real-scale records. Open the live app, the animated architecture and the guided playbook without leaving this page.
Open the live app inside this page.
The full Lovable front end — dashboard, compliance chat, customer portal and GDPR tooling — loads in a window over this page. No new tab required. Backed by a FastAPI service on Railway with a dual-persona API.
Why it matters
Governed AI, not a slideware demo.
I design and ship governed data + AI systems end-to-end — MDM-grade entity resolution and survivorship, lineage on every field, and a vector-RAG layer that answers compliance questions from trusted records, not raw text.
In regulated BFSI, a "single customer view" is only useful if every value is explainable and every AI answer is grounded. This shows that discipline as working software a CDO, steward or auditor can click through.
Governance controls
Pipeline
Ingest
Load all sources, parse, normalise — keep every record traceable.
Data Quality
Score completeness, format, consistency; quarantine weak records.
Entity Resolution
Fuzzy + Soundex/Metaphone + ML clustering group identities.
Survivorship
Trust weights, recency & regulatory lock pick winning values.
Golden Record
One authoritative profile with confidence and full lineage.
RAG Index
35k chunks in pgvector for grounded compliance queries.
Dual-persona API
Compliance (full) vs customer (filtered) self-service.
Live stats
Real scale, real evidence.
Stack