At Web Crypt Technology, I build generative and agentic AI solutions with LLM APIs integrated via AWS. I validate them against internal benchmarks and monitor run-level performance in LangSmith.
I also run EDA and feature engineering on structured and unstructured data, automate pipelines that clean records and flag issues, and train and evaluate ML models with experiments tracked in MLflow.
On my Autonomous AI Research & Task Execution Agent project, I built a supervisor-led multi-agent workflow that delegates work across Research, Data Analyst, and Report agents. It combines hybrid retrieval over uploaded documents with persistent memory and tools for web search, databases, and calculations.
I added a relevance cutoff to return “not found” rather than hallucinate, and measured 85% Recall@5 and an 82% tool success rate. I also developed an AI Customer Support Copilot that drafts RAG-grounded replies, routes uncertain conversations to human reviewers, and achieved 92% faithfulness in evaluation.
