Agentic credit analysis
Same-day credit decisions with agents that know when to stop
- Year
- 2024–26
- Client
- Experian
- Domain
- Banking · FINRA/SEC
- My role
- Lead AI/ML Engineer
- Stack
4h → 15mcredit analysis
95%hallucinations caught
98%compliance score
The problem
Analysts stitched together credit bureau data, market analytics and regulatory filings by hand, so loan decisions took days — and any AI help had to stay inside FINRA and SEC rules.
What I built
- Specialised retrieval agents for bureau data, market analytics and filings, coordinated by an orchestrator across 12+ microservices.
- Real-time tool calling so answers are built from live data, not model memory.
- Guardrails modelled on suitability rules, with every recommendation checked before it reaches a customer.
The hard part
Financial advice can’t be ‘mostly right’. I built a layered hallucination-detection pipeline that cross-checks claims against retrieved sources and blocks anything unsupported — it catches 95% of false statements.
Outcome
Credit analysis went from 4 hours to 15 minutes, enabling same-day decisions, with an 80% drop in complaints about inaccurate advice.
Client data and code are confidential; figures are from the delivered system.
Next caseClinical document digitisation →