Where product strategy meets technical rigor

Product leadership & AI engineering capabilities

Four years shipping AI-native B2B SaaS in regulated domains. The unglamorous parts: discovery, prioritization, regulatory sign-off, and killing the things that shouldn't ship.

4 yrs
Shipping AI-native SaaS
2
Companies, one thread
9
Documented case studies
GRC · FDA SaMD · Retail
Domains
01 - Grounded RAGEvery answer traces to a source, or says "N/A".
02 - Model tieringSmallest model that clears the bar, per task.
03 - Eval harnessesAccuracy measured before launch, not asserted after.
04 - HITL guardrailsA human signs off on anything regulated.
Key capabilities at a glance
Grounded RAG
Model Tiering
Eval Harnesses
HITL Guardrails
Domain breakdown

Deep-dive domains

Pillar 01

AI Product Strategy & Discovery

0→1 FramingMetric DesignKill Decisions
0→1 Discovery & Workflow Mapping

Shadow the team before proposing anything. A 41-minute structured walkthrough, not a stakeholder-interview template.

Methodology:Time-motion shadowing, five-whys root-cause, workflow diagrams before architecture.
Metric Design: North Star, Primary, Guardrail

Every scorecard has a guardrail metric that can go wrong, and it's reported honestly when it does.

Impact:Quote Parser's own guardrail caught a rollout regression: delivery time up 8.5%, reported as-is.
Scope Control & Kill Decisions

A well-argued recommendation against shipping is a deliverable, not a failure.

Discipline:$1M in engineering spend avoided by recommending termination in discovery, before sprint one.
Pillar 02

AI Systems Architecture & Prompt Engineering

Grounded RAGModel TieringEval Harnesses
Grounded RAG & Multi-Agent Pipelines

Five narrow agents beat one large prompt: each can fail, retry, and be tiered independently.

Benchmark:97% pipeline completion across 100 test sources; 20/20 reviewed citations grounded correctly, zero fabricated.
Model Tiering & Cost Modeling

Cost per unit is a product metric. If you can't state it, you can't price it.

Pricing:Haiku 4.5 for high-volume screening, Sonnet 4.5 for synthesis. $0.02–$0.09/assessment, $0.03–$0.06/quote.
Evaluation Harnesses

An accuracy claim without a labeled test set is an opinion.

Proof:Built a harness scoring extraction against hand-labeled ground truth: 100% recall/precision on 55 line items, 2 real quotes.
PROMPT DESIGN PRINCIPLE The prompt is where the domain lives: 7-step and 8-step fallback chains for pricing, written by someone who understands how analysts read a quote.
Pillar 03

Regulated Domains & Enterprise Governance

FDA SaMDHIPAAHealthcare GRC
FDA SaMD Compliance

A clinical MVP pivoted from speed-first to interpretability-first after clinicians said they wouldn't sign what they couldn't defend.

Result:0→1 clinical radiology AI, $500K+ ARR in 6 months, NLP accuracy 68% → 88%.
HIPAA & PHI/PII Risk

An uncited claim, or an unaddressed PHI leak, is worse than no claim at all.

Governance:Recommended termination of a $1M initiative after surfacing PHI leakage risk and negative ROI in discovery.
Healthcare GRC & Spend Analytics

A clinical AI POC for an enterprise hospital client, validated against their procurement data.

Value:$5–7M in annual savings identified for an enterprise hospital client.
Pillar 04

Executive Delivery & Operating Rhythm

WSJF / RICESAFe 5.0 POPMStakeholder Alignment
WSJF & RICE Roadmapping

Ship the narrow, highest-certainty slice first; let real usage tell you what's next.

Execution:Quote Parser's phased rollout WSJF-prioritized against 5 competing initiatives.
SAFe 5.0 POPM, Cross-Functional Delivery

Certified Product Owner / Product Owner under Scaled Agile.

Cadence:Ran weekly SLA triage and roadmap review across QA, Engineering, and Customer Success.
Stakeholder Alignment in Regulated Environments

Trade-offs in healthcare and GRC need defensibility, not just a demo.

Reach:Aligned Eng, QA, CS and Legal on competing priorities; presented roadmap trade-offs to senior leadership.
Strategic value proposition

How this differs from a conventional PM or an ML engineer.

Capability axisConventional PMML Engineer / ResearcherKrishnaS Hemanth
Discovery & Problem ScopingRelies on stakeholder interviews and roadmap templates.Starts from available data and model capability.Shadows the workflow first; a kill memo is a valid discovery output.
Architecture & QualityDelegates technical tradeoffs to engineering.Optimizes for benchmark accuracy alone.Writes the prompt rules directly; "N/A" beats a plausible guess.
Regulatory & ComplianceTreats compliance as a legal-review checkbox.Compliance is typically out of scope.FDA SaMD and HIPAA constraints shape the architecture from day one.
Execution & GovernanceMeasures success by feature velocity.Measures success by model metrics alone.Every number ships labeled Measured, Owner-reported, or Projected.
Production toolkit & stack

What I actually build with.

AI & Data Pipeline

Claude (Bedrock)RAG / pgvectorModel tieringEval harnessesPrompt designCost modeling

Engineering

Python (FastAPI)FlaskReactJavaScript.NET 8SQLite / Postgres

Product & Governance

WSJF / RICENorth Star metricsFDA SaMDHIPAAKill-memo artifactsStakeholder alignment

Execution & Tools

SAFe 5.0 POPMJIRAAzure DevOpsConfluencePendoFigma
Productivity
+11.7%
Quotes per analyst / mo
147.0 vs 131.6 in 2025
Revenue
$500K+
ARR in 6 months
0→1 clinical radiology AI
Savings
$5–7M
Annual savings identified
Enterprise hospital client
Judgment
$1M
Spend killed on purpose
PHI/PII risk + negative ROI

Have a high-stakes AI or enterprise product challenge?

Available for AI product strategy advisory, clinical compliance architecture reviews, and 0→1 enterprise execution.

Schedule a conversation LinkedIn →