Headline

AI Product Manager |GenAI (LLM, RAG, Agents) | Top PM Fellow @NextLeap |Ex-Founder | COEP

About

I build AI systems and own the product problem behind them — hybrid RAG, multi-agent orchestration, human-in-the-loop workflows, and the eval sets that gate them. Before that I founded a diabetes healthtech venture: ran a 50-patient clinical trial, and killed a related AI prediction project after concluding it couldn’t be built safely. My instinct is to use the lowest level of AI that solves the problem and put deterministic controls around it wherever the output matters.

Location

Bengaluru

AI stack

anthropic apis

antigravity

claude code

claude

codex

cursor

Work experience

AI Product Management & Applied GenAI Fellow at NextLeap, Jul 2025 - May 2026 · 11 mos, Built 7+ end-to-end AI products as fellowship projects — problem framing to deployed MVP. ▸ Designed RAG + multi-agent systems covering retrieval, model selection, orchestration, and evaluation ▸ Deepest build: 3-pillar advisory platform on live traffic — hybrid RAG, 7-state voice agent, human approval gate. Eval suite gates every release: 35 golden + 16 adversarial prompts at 100% required refusal ▸ Made explicit GPT-4o vs Claude vs OSS decisions based on cost/latency/accuracy tradeoffs ▸ Built LLM eval frameworks — golden datasets, A/B testing, rubric-based scoring — across 3+ projects ▸ Tracked cost-per-query and P90 latency as core product KPIs ▸ Defined PRDs, roadmaps, L0/L1/L2 metrics for every build

Founder & AI Product Lead at QYOUR, May 2023 - May 2025 · 2 yrs 1 mo, Pune, Maharashtra, India, Founded a natural diabetes supplement business — built an AI-powered website as the purchase conversion funnel. ▸ Identified the core barrier: users needed trust before buying — built an AI funnel (health quiz, chatbot, glucose tracker, meal suggester), each feature targeting a specific user anxiety ▸ Segmented to newly diagnosed Type 2 diabetics, HbA1c 5.7–7.5% — explicitly excluding severe cases needing medical intervention ▸ Clinical trial: 50 patients, 12 weeks, single-arm — 0.9% HbA1c reduction, 92% completion, side effects <5% ▸ Set the North Star as a health outcome, not engagement: 60%+ of users achieving ≥0.5% HbA1c reduction ▸ Closed pre-launch: paid a manufacturer who never delivered product. I validated the science and the segment but never stress-tested single-supplier dependency ▸ Ran a nine-month feasibility assessment on AI diabetes prediction — concluded consumer-accessible data couldn’t reach the required accuracy or safety threshold, and recommended stopping

Founding Member : Business & Strategy at Aimshilp Earth industries, Apr 2020 - Jun 2023 · 3 yrs 3 mos, Mumbai, Scoped a bamboo-for-timber plywood venture in Ratnagiri after the 2017 Bamboo Amendment opened non-forest cultivation — market sizing, unit economics, and species validation with the National Bamboo Mission. Stopped at planning stage with no offtake or proven buyer. The lesson — validate demand before building supply — I now apply to every product decision.

Management Trainee at Lenskart.com, Oct 2018 - Mar 2019 · 6 mos, Mumbai, Maharashtra, India, Improved NPS and fulfillment efficiency via 100+ customer interaction analysis

Management Trainee at Reliance Energy, May 2017 - Jul 2017 · 3 mos, Mumbai, Maharashtra, India, Designing, implementing, and maintaining SCADA systems for electrical infrastructure, such as power grids, transmission lines, and distribution systems

Education

Applied Generative AI, Artificial Intelligence, Nextleap

AI Product Management, Nextleap

Public links

github: https://github.com/aartipm2025-max

medium: https://medium.com/@aartipm2025

nextleap.app: https://nextleap.app/portfolio/aarti-dhavare