Projects.

Problem

Users struggled to find relevant datasets using keyword search. Queries often returned too many irrelevant results or missed useful tables entirely, leading to low trust in search and frequent reliance on tribal knowledge.

Context

Enterprise data platform on GCP used by analysts and data teams across multiple business units. Search was a critical workflow entry point but ranked low in satisfaction and effectiveness.

My Role

Product Manager
Owned discovery, roadmap, and launch of AI-powered search capabilities. Worked closely with engineering, data science, data platform, and UX teams to define scope, iterate on prompts, and evaluate results.

Solution

Built a RAG-based semantic search system using Gemini Pro:

• • Implemented vector embeddings for tables and metadata
• • Designed hybrid search combining semantic + keyword retrieval
• • Used Gemini Pro to generate contextual relevance and summaries
• • Created prompt framework to improve result interpretation
• • Prioritized top-used datasets for initial rollout
• • Built evaluation loop using real analyst queries

Focused on usefulness over novelty — ensuring results helped users complete real analysis tasks.

Impact

40% fewer failed searches · higher trust in search as entry point · foundation for AI discovery

• • Improved relevance of search results for top datasets
• • Reduced friction in locating commonly used tables
• • Increased confidence in search as a starting point for analysis
• • Enabled more natural-language queries across the platform
• • Established foundation for future AI-driven discovery features

Tech Stack

Key Learnings

• → AI search must be grounded in real workflows, not demos
• → Hybrid retrieval consistently outperformed pure semantic approaches
• → Metadata quality is as important as model quality
• → Evaluation with real queries is critical for trust
• → Users care more about usefulness than novelty

Overview

Led product for the Lumi web portal during migration of a large enterprise big data platform from on-prem infrastructure to GCP. Owned discovery, project creation, homepage UX, and workflow automation to reduce friction and drive adoption across analysts and data teams in 12+ business units.

My Role

Product Manager
Owned roadmap and execution across search, onboarding, and automation initiatives. Partnered with engineering, design, data, and governance teams to deliver incremental improvements that supported cloud migration and scaled usage across the organization.

What I Delivered

Impact

Cut onboarding time by 40%, reduced manual effort by 60% through automation, and scaled platform adoption to 30K+ users across 12+ business units.

• • −40% onboarding time through automated project setup
• • −60% manual effort via workflow automation
• • Scaled platform adoption to 30K+ users across 12+ business units
• • Improved discovery and reduced reliance on tribal knowledge
• • Established foundation for AI-powered search and recommendations

Tech & Environment

Overview

Built an LLM-powered feedback analysis tool to cluster, summarize, and prioritize product feedback. Implemented scoring across frequency, sentiment, and impact to surface high-value roadmap insights from large feedback datasets.

Problem

Product teams received feedback from multiple sources but lacked a systematic way to synthesize themes, prioritize issues, and identify patterns across large volumes of input. Manual analysis was time-consuming (hours per analysis session) and prone to missing important signals.

Solution

Created an LLM-powered pipeline that:

• • Clusters similar feedback items using semantic similarity
• • Generates concise summaries for each cluster
• • Scores themes based on frequency, sentiment, and business impact
• • Surfaces high-priority issues for roadmap planning
• • Provides actionable insights with supporting evidence

Built as a rapid prototype to demonstrate the value of AI-assisted product decisions and validate the approach with real feedback data.

Impact

Cut feedback analysis from hours → minutes · enabled rapid AI prototyping and decision-ready insights

• • Reduced analysis time from hours to minutes (10x+ improvement)
• • Demonstrated rapid AI prototyping from idea to working product
• • Enabled decision-ready insights from large feedback datasets
• • Validated approach for AI-assisted product prioritization
• • Improved visibility into recurring user pain points

Demo Video

Watch a walkthrough of the feedback analysis tool in action

Tech Stack

Key Learnings

• → Rapid prototyping helps validate AI use cases before large investments
• → Scoring logic needs to balance multiple signals (frequency, impact, sentiment)
• → Cluster quality depends heavily on good prompt engineering
• → Teams need explainability — showing supporting evidence builds trust
• → AI works best augmenting PM judgment, not replacing it