Senior Product Manager — Agentic AI & Generative AI
10+ years building and shipping enterprise AI, ML, and data products across financial services and platform SaaS. I build working prototypes to validate AI product ideas — not to prove I can write code, but because the fastest way to understand what a product should do is to make it do something real.
AI-native tools that solve operational problems knowledge workers face every day — meeting ops, product spec writing, sales intelligence, supply chain automation. Each project is a full-stack working prototype: backend API, agent orchestration, and a deployed frontend.
AI Meeting Co-pilot — 🟢 Live demo
Agentic system that turns a meeting transcript or audio file into structured action items, routed tasks (Jira/Asana/Linear), a follow-up calendar event, and a formatted summary email. Built with LangGraph (7-node stateful graph), Claude API tool use, OpenAI Whisper, and Instructor + Pydantic for typed outputs.
PM Spec Generator — 🟢 Live demo
AI pipeline that transforms a 2-sentence feature idea into a developer-ready PRD with live competitive research, typed user stories, Given/When/Then acceptance criteria, success metrics, and edge cases. Built with Claude API web search tool use, Instructor + Pydantic, FastAPI SSE streaming, and React.
AI Knowledge Copilot — 🟢 Live demo
Document intelligence assistant that ingests PDFs, DOCX, PPTX, and CSV files and enables natural language Q&A, summarization, and multi-document comparison. Built with OpenAI GPT-4o-mini, RAG-lite architecture, and Streamlit.
AI sales intelligence tool for Enterprise SDRs. Aggregates prospect signals across financial reports, LinkedIn activity, and tech stack data to generate personalized, brand-compliant outreach. Built with TypeScript.
Autonomous Supply Chain Digital Worker — 🟢 Live demo
Agentic AI prototype simulating how enterprise digital workers detect, analyze, plan, and execute actions across supply chain workflows. Multi-agent orchestration with real-time decision routing.
| Layer | Tools |
|---|---|
| Agent orchestration | LangGraph, LangChain LCEL, multi-agent graphs, conditional edges |
| LLM APIs | Claude API (Anthropic), OpenAI GPT-4o, tool use, extended thinking |
| Structured output | Instructor, Pydantic v2, JSON schema enforcement |
| RAG & memory | Pinecone, FAISS, text-embedding-3-small, cosine similarity, episodic memory |
| Speech-to-text | OpenAI Whisper, speaker diarization |
| Backend | FastAPI, Uvicorn, SSE streaming, background tasks |
| Frontend | React 18, Vite, Streamlit |
| Prompt engineering | XML-structured prompts, few-shot examples, chain-of-thought, tool use loops |
| Observability | LangSmith — token cost, latency, agent trace |
| Deployment | Render (FastAPI), Vercel (React), Streamlit Cloud |
I came to AI product management through a technical path — engineering and data science before product, then 10+ years leading AI and ML product work in enterprise environments. Financial services and platform SaaS. I've built pricing models, fraud detection systems, data platforms, and now agentic AI applications.
The prototypes here are how I think through product problems: build a working version, find where it breaks, understand the tradeoffs, then write the spec.