AI PM who shipped 6 GenAI products in 8 months — consumer funnel to voice to computer vision
At Stanza Living I own the website funnel driving 50% of 40,000 monthly leads, shipped Riya (a Hinglish voice bot on GPT-4.1), and built computer-vision compliance across 22 clusters. At Power Finance Corporation I built NLP-driven risk detection on a ₹40,000 Cr portfolio. IIM Indore MBA.
Career Timeline
Shipped 6 production AI systems in 8 months. Consumer funnel, voice AI, computer vision compliance, generative comms, AI video pipeline, NL2SQL copilot.
NLP-driven early-risk detection on a ₹40,000 Cr infrastructure loan portfolio. Regulated data environment with PII masking and role-based access.
PM Case Competition finalist (Tekion 2022, WITDA 2022). PSPO II, CSM certified.
Built an adaptive-learning platform for competitive-exam prep. Led 11-person team, MVP to go-to-market. Scaled revenue 50%, grew engagement 40%.
NTSE Scholar — top 0.1% India.
Systems I've Shipped
Production AI that moved real metrics — not slide decks.
Consumer Funnel Ownership
Own the website driving 50% of Stanza's 40,000 monthly leads
Problem
No single PM owner on the seeker-side website. Organic share was flat, funnel had high drop-off at key conversion points, and there was no structured experimentation culture.
Solution
Took end-to-end ownership of the Stanza website as growth PM. Ran SEO experiments, CRO interventions, and UX redesigns across the full acquisition funnel. Built a structured A/B testing cadence.
Impact
Riya: AI Voice Bot for Lead Conversion
Hinglish GPT-4.1 bot handling lead visit confirmation and nudging at scale
Problem
Lead visit confirmation was manual — agents spent time on routine nudge calls. English-only bots failed on the Hinglish code-switching common to Stanza's audience. No fallback intelligence for low-confidence calls.
Solution
Shipped Riya on GPT-4.1 + Cartesia Sonic 3 with live language-drift tracking. Added confidence-based human handoff: negative sentiment or 2-turn context loss routes to human agents automatically.
Impact
Computer Vision Housekeeping Compliance
Cut inspection turnaround from 12 hours to 15 minutes across 22 clusters
Problem
Housekeeping audits were manual across 22 operational clusters with 400+ stakeholders. The 12-hour inspection cycle created a full-day lag in compliance visibility and left audit coverage gaps.
Solution
Led VLM-based inspection system on AWS CPU inference. Set 65% confidence threshold — below it routes to human review. Wired drift monitoring and retraining triggers to prevent silent degradation.
Impact
NLP-Driven Credit Risk Intelligence
Automated 85% of manual financial-data parsing on a ₹40,000 Cr portfolio
Problem
Credit analysts spent days each month manually compiling risk signals from unstructured documents across a ₹40,000 Cr infrastructure loan portfolio. Legacy data silos made early-signal detection near-impossible.
Solution
Built entity-extraction and heuristic-scoring models to parse unstructured credit data and flag early-risk signals. Shipped portfolio-intelligence dashboards with PII masking, role-based access control, and data minimisation for the regulated environment.
Impact
How I Ship AI Products
Four principles behind every system I build.
Evaluation First
Before shipping, I define eval sets, run LLM-as-judge scoring, and set precision/recall thresholds. Nothing goes to prod without a passing baseline.
Confidence Routing
Every AI system I ship has a threshold below which work routes to humans. The 65% VLM gate and Riya's sentiment-based handoff are both live examples of this principle.
Guardrails as Architecture
Tone classifiers, PII masking, and adversarial testing aren't QA steps — they're in the design from day one. Trust is architecture, not a feature.
Drift Monitoring
AI systems degrade silently. I wire retraining triggers and performance alerts into every deployment so the team knows before users do.
Early Builds
Prototypes from 2022–23 that shaped how I think about evaluation and user feedback before applying those lessons at Stanza.
MoodBite
Voice-first AI food concierge that understands mood, craving, and context to recommend the right meal.
View Project
FeatureFit
AI-powered feature prioritization tool for PMs. Upload your backlog, get RICE-scored recommendations.
View Project
LLM Tuner
Prompt experimentation playground for testing, comparing, and iterating on LLM outputs side-by-side.
View ProjectStrategic Writing
AI Products Fail on Power, Not Models
Why the biggest risk in AI product development isn't model quality — it's organizational power dynamics.
LLM-EO: The Future of Search
How LLM Engine Optimization is reshaping the way content gets discovered and consumed online.
Designing for LLM-EO
A practical framework for making your product visible to LLM-powered search engines.
Looking for an AI PM who ships full-stack LLM products?
From consumer funnels to voice AI to computer vision — I build and own the product end to end. I'm in market for Senior AI PM roles (40-60 LPA) with a 3-week timeline.