HV

Co-founder · Forward-deployed engineering

Harsh Vitra

Harsh builds and ships zero-to-one AI products with Brandlabs — architecture, evals, and production systems inside client environments. He focuses on making regulated AI measurable and operable, not demo-grade.

  • Forward-deployed AI engineering
  • LLM evaluation in production
  • Zero-to-PMF delivery

Writing

Choosing an LLM vendor under regulated constraints.

BAA availability, retention defaults, region options, and eval portability — how Brandlabs shortlists model providers for healthcare and regulated buyers.

Safety evals for clinical and health chatbots.

How to evaluate refusal, escalation, medical-advice boundaries, and hallucinated dosages — before a health assistant meets real users.

Building a prompt regression dataset that ages well.

Golden sets, adversarial sets, and ownership so your LLM suite still means something after six months of prompt thrash.

Three eval dimensions: functional, qualitative, safety.

The Brandlabs Discovery Sprint pattern for setting LLM eval targets in weeks 1–3 — before feature thrash eats the timeline.

CI that blocks deploys on failing evals.

Sampling, cost caps, flake handling, and which LLM eval suites must stay required vs nightly — so production cannot silently regress.

The two-week Discovery Sprint, unbundled.

Inputs, outputs, kill criteria, and how Brandlabs maps a Discovery Sprint into a 12–20 week zero-to-PMF engagement.

Zero to PMF in 12–20 weeks: the phase model.

Foundation → build to alpha → production-harden → PMF hunt — with eval and compliance gates at each phase.

Safety engines for WhatsApp health assistants.

Escalation, localization, and content safety patterns for healthcare chat on WhatsApp — from shipping Myna Bolo in production.

AI for Indian MSME compliance (lessons from Correct).

Productizing 69,000+ obligations and regulations that change ~15 times a day — without turning an AI assistant into unauthorized legal advice.

Evals are the product.

Why we treat the eval suite as a first-class deliverable — and how it changes how we negotiate scope.

Every postmortem becomes a regression test.

The cultural and technical loop that turns production incidents into permanent guardrails.

What a forward-deployed engineer actually does.

Not a project manager, not a consultant. A senior engineer who lives in your Slack and writes the hardest code.