AI & Data Product Leader · Product, Evaluation & Quality (AI/ML) · Conversational & Agentic AI
AI/Data product leader, 10+ years building data & GenAI products end-to-end — data product lead for Walmart's AI shopping assistant, owning the evaluation, experimentation, and quality systems that steer the roadmap.
LinkedIn · Streamlit · Tableau Public · RPubs · Email
I take customer-facing AI products from ambiguous problem to launch, and own the evaluation, experimentation, and quality systems that decide what ships next. My edge is hands-on technical depth — LLM evaluation, RAG, observability, experimentation infrastructure — paired with the product judgment to weigh customer experience, safety, cost, and scale in one call.
Currently data product lead for Sparky, Walmart's AI shopping assistant (used by ~50% of Walmart app users; a publicly cited driver of ~35% larger orders), where I defined the platform's first standardized quality KPI and its greenfield evaluation standards from zero.
Five runnable apps spanning the AI-product lifecycle — build → evaluate → experiment → monitor → explain. Each is live, self-contained, and built on synthetic or real public data.
| Project | What it shows | Demo |
|---|---|---|
| 🛰️ LLM Observability & Evals | Model-health monitoring across quality, safety, performance, cost & drift — SQL-backed pipeline, alerting, and PDF/PPTX export | Live ↗ |
| 💬 Chat Quality Score (CQS) | LLM-as-a-judge evaluation scoring conversations on a 4-dimension rubric, calibrated against human labels | Live ↗ |
| 🛒 Product Recommendation Quality | Tracks AI recommendation relevance week over week and surfaces the drivers behind any change | Live ↗ |
| 🧪 A/B Experimentation Framework | Hypothesis design, randomization, guardrail metrics, and ship / iterate / stop decisioning | Live ↗ |
| 🔎 LedgerIQ — Finance RAG Agent | Finance-ops RAG over two sources — real SEC EDGAR filings and FP&A planning documents — grounded, cited answers that refuse when out-of-corpus, with token-minimization controls and MCP retrieval servers | Live ↗ |
LedgerIQ runs on real public SEC EDGAR data (SEC source) plus synthetic FP&A documents (FP&A source); the other apps use fabricated or synthetic data — no proprietary, confidential, or employer-specific information.
Built with Streamlit · RAG & MCP · SQLite · LLM-as-a-judge · Python
Product: product strategy & roadmap · feature prioritization · MVP scoping · PRDs & requirements · experimentation & A/B testing · KPI ownership · stakeholder management GenAI & AI/ML: LLM evaluation (LLM-as-a-judge) · RAG & grounding · agentic AI & tool use (MCP) · prompt evaluation · conversational & agentic AI · retrieval / recommendation relevance · human-in-the-loop governance · model observability · AI safety evaluation · token & cost–quality optimization Data & Platform: SQL · Python · R · BigQuery · Snowflake · PostgreSQL · Kafka · telemetry & experimentation infrastructure BI & Tools: Tableau · Power BI · Streamlit · Jira · Miro
- Bravo Award (×2) — for GenAI initiatives delivering ~$1M in annual savings, and for analytics spanning 30+ conversational-AI domains
- Innovation Challenge Winner — top RPA solution selected from 218 ideas across 340 professionals, funded and rolled out across the US, Europe, and India
Open to AI/GenAI Product Management roles. Let's talk → chowdhuryanimesh1@gmail.com