All projects
Multi-agent LLM pipelineTechClan, Lahore · Venture capital
DealTitan and Pitch Deck Analysis
AI due diligence for investors and founders: a multi-agent LangGraph pipeline that turns a pitch deck into an expert-grade report in about 5 minutes, at about $5 per deck.
Two paid products on one Django and Celery backend with a React and TypeScript front end. The LangGraph graph runs research, extraction, forensic checks, section analysis and synthesis over Gemini and Claude. My work covered new pipeline steps, deck screening and duplicate blocking, credit settlement with credit-neutral retries, per-call cost reconciliation and prompt caching, investor thesis-fit scoring, founder interview practice and deck versioning, plus TOTP MFA, session management and the multi-stage release process behind it.
Highlights
- Extended a multi-agent LangGraph pipeline that has analyzed 1,100+ pitch decks, cutting expert review from hours or days to about 5 minutes at about $5 per deck (95%+ faster).
- Eliminated wasted LLM spend with mandatory deck screening and duplicate-upload blocking, saving a $5 run and a customer credit per bad upload.
- Led credit settlement and failure recovery, adding credit-neutral retries that re-run only a failed section instead of the full report.
- Reconciled per-call LLM costs and introduced prompt caching on Claude calls for accurate, lower per-report cost.
- Delivered investor thesis-fit scoring, showing whether each deal fits an investor's mandate across inbox, reports and CRM.
- Launched founder AI interview practice with feedback grounded in real answers, plus deck versioning to track improvements.