I build
AI agents that
ship to production.

A US-based logistics MNC's procurement team was manually validating thousands of supplier records across fragmented sources — tax IDs, addresses, banking details, compliance fields. Onboarding cycles were slow and error-prone.

Built a custom validation agent with n8n + OpenAI + OCR that autonomously cross-references supplier-submitted data against authoritative sources, flags discrepancies, and produces decision-ready outputs for human reviewers.

Enterprise clients needed rich, real-time company profiles — firmographics, leadership, risk signals — but legacy data vendors were expensive, stale, and incomplete.

Built a 7-module intelligence engine that aggregates data from public registries, news, web sources, and proprietary APIs, uses LLM agents to normalize and validate entities, and delivers structured, decision-ready company profiles on demand.

Enterprise procurement platforms faced fraudulent supplier onboarding — shell companies, identity mismatches, and forged documents slipping past manual KYC and creating financial and reputational risk.

Designed a hybrid ML + LLM fraud detection agent that scores supplier submissions across dozens of signals — corporate registry checks, document authenticity, behavioral patterns, and public data anomalies — flagging high-risk applications in real time.

A US energy utility received thousands of Certificates of Insurance (COIs) in inconsistent formats — PDFs, scans, photos. Manually verifying coverage limits, expiration dates, and named insureds created a persistent compliance bottleneck.

Built an IDP agent combining OCR, LLM extraction, and validation logic that parses COIs, extracts structured coverage data, validates against policy requirements, and flags expiring or non-compliant certs automatically.

Support teams were receiving high volumes of customer tickets describing product issues in unstructured, inconsistent language. Triaging, classifying, and routing each ticket to the right engineering team was slow, which meant customers waited longer and engineers received vague bug reports lacking the context needed to debug quickly.

Built an LLM pipeline that reads incoming support tickets, classifies the underlying product defect category, extracts reproduction steps and severity signals, and auto-generates a structured engineering ticket with all the context developers need — closing the loop between customer pain and engineering action.

Business teams depended on analysts and engineers for every ad-hoc data question — revenue queries, funnel drop-offs, anomaly investigations. The bottleneck slowed decision-making and buried data teams in repetitive reporting work.

Built an AI analytics copilot combining text-to-SQL generation, conversational data Q&A, scheduled automated reporting, and anomaly detection — grounded in the company's data warehouse schema so non-technical users can ask questions in plain English and get trustworthy, cited answers.