name:          Mahesh Kumar
current_role:  Lead Product Owner, GenAI Products @ RBC Insurance
education:     Master of Management in AI — Queen's University Smith School of Business
background:    17 years in Canadian financial services (RBC, CIBC Enterprise Data Hub)
location:      Mississauga, Ontario, Canada
approach:      Ship first. Document everything. Explain every decision.
building:      AI products in regulated Canadian domains — in public

I build AI products that solve real consumer problems in regulated domains — insurance, healthcare, and travel. My background is in financial services product management, not software engineering. Everything here is built by a PM who learned to build in public.

Built in 2026. Publicly accessible. Architecturally documented. No proprietary data.

Domain: Personal Insurance  |  Stack: RAG + Claude API + Lovable

Plain-language insurance policy explainer. Upload your PDF, ask anything, every answer cites the exact clause and page. Built because FCAC data shows 52% of Canadians find their insurance policy hard to understand — and 23% have never read it.

Domain: Travel / RegTech  |  Stack: Multi-Agent + Condition Framework + Claude API + Lovable

Neutral ADR readiness platform for Canadian air passenger disputes (APPR). Triggered by the Air Canada ADR pilot announcement, April 8, 2026. Works for Air Canada, WestJet, Air Transat, Porter — any Canadian airline.

Key product decision: Removed XGBoost ML probability scores because predicting legal arbitration outcomes is unreliable and harmful. Replaced with a transparent 6-factor APPR condition checker — each factor mapped to SOR-2019-150.

APPR Rule Engine (38/38 tests) → Agent 1: Evidence Assessor (Claude API)
→ Claim Readiness Assessment (6 APPR factors, no ML scores)
→ Agent 3: Submission Drafter → Agent 4: ADR vs CTA Guide

Domain: Responsible AI / Banking RegTech  |  Stack: FastAPI + FAISS + Claude API

Compliance agent for Canadian financial institutions navigating overlapping AI governance frameworks — PIPEDA/Bill C-27, OSFI model risk guidance, AIRA, and internal AI policies. Built as a portfolio piece targeting Director-level AI Enablement roles at Canadian banks.

Domain: Healthcare  |  Stack: Multi-Agent + ML + Claude Code

Plain-language care navigation for Canadians. "Where should I go and how urgently?" Four urgency levels with a hard 911 override for red flag symptoms. Ontario and BC coverage in v1. Context: 6.5M Canadians without a family doctor (CMA 2026), 30-week average specialist wait.

Queen's MM in AI coursework, experiments, and domain explorations across insurance, healthcare, and analytics.

1. Ship the rule engine before the LLM — deterministic logic always before generative. If the answer is certain, it should not hallucinate.

2. Design for failure first — every PM brief has a failure modes section before the feature list. The failure modes shape the architecture.

3. Transparency over performance — ClaimReady AI's most important decision was removing a well-performing ML model because showing probability scores in a legal context creates harmful expectations. Honest and useful beats impressive.

4. Every agent earns its place — multi-agent architectures are only as good as the reason each agent exists independently.

5. Domain depth is the moat — 17 years in regulated Canadian financial services means I know which AI product decisions actually matter in production, not just in demos.

AI / ML

Build & Deploy

Regulated domains

• 🔨 Building CareNav AI — healthcare triage, Ontario + BC
• ✍️ Writing AI PM in the Wild — weekly newsletter on AI products in regulated domains
• 🎯 Targeting Director-level AI Enablement roles at Canadian financial institutions
• 📦 Publishing the AI PM Portfolio Kit — PRD templates, PM briefs, architecture frameworks from all four builds