Three roles. One thread.
Each job taught me a different layer of the same problem: how to take something messy — user behaviour, an LLM's worst day, a vague product brief — and turn it into something a team can act on.
EMA
AI Evaluation Analyst · Agentic AI for the enterprise
EMA builds AI agents for the enterprise — sales, support, HR, the long tail of knowledge work. My job is to make sure those agents are trustworthy enough to ship. When I joined, there was no structured evaluation process for the 50+ agents in the pipeline. I built it.
Built an internal full-stack app in Claude Code to fix the fact that nobody had a single view of live deals. Designed the backend workflows for AI-powered data extraction and document upload, demoed to leadership, and shipped it. Now used by 20+ folks across GTM and Solution Architecture.
Designed and owned the end-to-end QA framework for the 50+ AI agents — defined evaluation criteria, set up test harnesses, ran regression sweeps. Directly unblocked the first public launch and supported closure of a key enterprise deal.
Diagnosed consistent failure patterns in LLM outputs causing customer escalations. Worked with engineering on targeted fixes — prompt tweaks, retrieval changes, agent routing logic. Improved accuracy by 20–30%.
Led rollout and validation of 50+ third-party integrations where production reliability had been inconsistent. Cut the issue rate sharply and improved end-user reliability at scale.
Surfaced evaluation data as structured input to PM prioritisation. Directly influenced feature tradeoff calls across multiple release cycles — including a few "no, don't ship that yet" calls.
Identified recurring AI output failures before they became widespread complaints. Prioritised fixes that reduced user-facing errors and improved retention.
Phenom
Product Analyst · Talent experience platform
Phenom is an enterprise talent platform serving Fortune 500s. I sat at the intersection of CS and Product — close enough to user pain to see the early signals, close enough to engineering to actually do something about them.
Identified behavioural churn signals nobody was actively monitoring. Built an alert system for the CS team. Prevented an estimated $200K in revenue loss.
Analysed drop-off patterns across user journeys, identified friction points, drove targeted fixes. Result: a 5% lift in feature adoption on the touched flows.
Defined and tracked 35+ metrics across retention and engagement where no unified KPI framework existed. Built leadership a consistent dashboard for data-backed decisions.
Translated raw behavioural data into weekly executive reports that directly shaped roadmap priorities — and surfaced issues early enough to fix them before the next QBR.
Blink X · JM Financial
Product Management Intern · Discount brokerage
Blink X is JM Financial's retail trading product. I joined as a PM intern — partly to ship a new lending product, partly to take a swing at a stock education app the team was sketching.
Analysed 25M+ user records to identify high-value personas and define the GTM targeting strategy for a new lending product.
Took full ownership of a stock education app from concept to prototype. Defined the roadmap, designed all UX flows, integrated gamification mechanics.
Conducted structured competitive analysis across 10+ players in the discount brokerage and edu-tech space. Identified positioning gaps and sharpened differentiation.
Where the thinking got shaped.
Plaksha University
Tech Leaders Fellowship · PG Programme in AI & Leadership
Selected as 1 of 60 young leaders from India on a $6,000 merit scholarship — a 6% selection rate. Run in collaboration with UC Berkeley and Purdue.
Ambedkar Institute of Technology
Bachelor of Computer Applications
The foundation — algorithms, web dev, databases, and the slow realisation that the most interesting questions were never about the code.