I build AI products
people can actually use.

ProblemRecruiting is slow and manual — sourcing candidates, screening, and running first-round interviews eats a team's week before anyone talks to a strong fit.

BuiltAn end-to-end AI recruitment platform: define a role from a job description, and it drafts a hiring funnel with stage-specific voice agents, sources candidates across multiple providers, runs voice-based screening interviews, and generates assessments to guide recruiter decisions.

My partBuilt the platform end-to-end — the funnel builder with per-stage voice agents, a candidate pipeline with bulk CSV import and de-duplication, people-search across four provider integrations (Apollo, PDL, Proxycurl, Coresignal), and 'evidence by construction' where every call result is stored for recruiter review with audit logging.

ProblemMeeting notes get lost. Decisions, commitments, and open questions vanish into hours of recordings nobody rewatches.

BuiltA live, fully usable platform (real sign-up, not a demo) that turns meetings into searchable memory — capturing decisions, commitments, and open questions with exact supporting quotes across Google Meet, Zoom, and Teams.

My partBuilt end-to-end (0→1): the capture pipeline, a RAG chat layer to query meeting history, auto-generated pre-meeting briefs and follow-ups, and multilingual transcription across 22+ Indian languages including mixed-language calls.

ProblemPharma teams track clinical-trial data, industry news, and SEC filings across scattered sources with no single grounded, citable view.

BuiltA pharma-intelligence platform that aggregates clinical-trial data, industry news, and SEC regulatory filings into one searchable RAG interface with sub-2-second query latency.

My partDesigned a two-layer architecture: a Celery ingestion pipeline feeding hybrid dense + sparse retrieval with RRF fusion and cross-encoder reranking over pgvector, plus a deterministic entity spine for grounded, cited answers.

ProblemLoan and finance docs are dense; support teams need accurate, source-grounded answers, not guesses.

BuiltA finance-domain RAG assistant with PDF ingestion, embeddings, and vector search for accurate, source-grounded answers.

My partBuilt the ingestion, embedding, and retrieval pipeline end-to-end with LangChain and Pinecone.