Field notes on health AI safety, privacy engineering and evals.https://www.brandlabs.app/enhttps://www.brandlabs.app/blog/phi-minimization-rag/https://www.brandlabs.app/blog/phi-minimization-rag/RAG for healthcare fails compliance when every chunk is treated as fair game for the prompt. Brandlabs designs retrieval with minimization: strip identifiers before embed where possible, filter by role and purpose, refuse when evidence is thin, and eval for leakage — not only answer quality.Thu, 30 Jul 2026 00:00:00 GMTRegulated AI architecturecontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/dpdp-consent-ux/https://www.brandlabs.app/blog/dpdp-consent-ux/For Indian buyers, DPDP-aware AI products need clear notice, purpose-bound processing, and consent UX users can understand — designed into onboarding and model features, not bolted on as a PDF. Brandlabs treats consent as a product surface with auditability, not a checkbox theater.Wed, 29 Jul 2026 00:00:00 GMTRegulated AI architecturecontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/baa-before-phi/https://www.brandlabs.app/blog/baa-before-phi/Do not put real PHI into an LLM path until business associate agreements, environment controls, and retention rules exist for every vendor that can see that data. Prototypes can move fast on synthetic or properly de-identified data. Crossing the PHI line early is how teams inherit avoidable risk.Tue, 28 Jul 2026 00:00:00 GMTRegulated AI architecturecontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/hipaa-vs-dpdp-posture/https://www.brandlabs.app/blog/hipaa-vs-dpdp-posture/Do not clone a HIPAA checklist onto an India launch or a DPDP notice onto a US clinical product. Build a shared spine — minimization, access control, auditability, vendor control — then layer HIPAA BA/safeguard sequencing for US healthcare and DPDP notice/purpose/consent for Indian personal data.Mon, 27 Jul 2026 00:00:00 GMTRegulated AI architecturecontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/model-vendor-risk/https://www.brandlabs.app/blog/model-vendor-risk/Pick LLM vendors on contract posture and control surfaces first, model leaderboard second. For regulated AI, Brandlabs scores BAA/DPA readiness, retention and training defaults, regional processing, logging, and how hard it is to swap providers when evals say the model drifted.Sun, 26 Jul 2026 00:00:00 GMTRegulated AI architecturecontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/audit-logs-for-llm-calls/https://www.brandlabs.app/blog/audit-logs-for-llm-calls/Every inference that can touch regulated data needs an audit event: who triggered it, when, which model/version, what tools ran, and whether it succeeded — without writing raw PHI into the log by default. Retention and access to that trail are product decisions, not afterthoughts.Sat, 25 Jul 2026 00:00:00 GMTRegulated AI architecturecontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/safety-evals-clinical/https://www.brandlabs.app/blog/safety-evals-clinical/Health and clinical assistants need a dedicated safety suite: out-of-scope medical advice, emergency escalation, dosage and diagnosis hallucination, jailbreaks, and empty-evidence refusal. Qualitative fluency is not safety. Brandlabs puts these fixtures in the required CI gate for regulated health products.Fri, 24 Jul 2026 00:00:00 GMTEvals & production AI qualitycontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/prompt-regression-datasets/https://www.brandlabs.app/blog/prompt-regression-datasets/A regression dataset dies when nobody owns it. Brandlabs splits golden journeys, adversarial safety cases, and production-derived fixtures — with labels, owners, and deletion rules — so CI stays trustworthy instead of becoming a junk drawer.Thu, 23 Jul 2026 00:00:00 GMTEvals & production AI qualitycontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/eval-dimensions-functional-qualitative-safety/https://www.brandlabs.app/blog/eval-dimensions-functional-qualitative-safety/Score AI products on three dimensions from day one: functional correctness, qualitative usefulness, and safety/compliance behavior. Set numeric targets in the Discovery Sprint so Friday demos and CI gates share one definition of done.Wed, 22 Jul 2026 00:00:00 GMTEvals & production AI qualitycontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/ci-evals-every-commit/https://www.brandlabs.app/blog/ci-evals-every-commit/Treat critical eval suites like unit tests: they run on every commit or deploy pipeline and fail the build when thresholds break. Keep a larger nightly suite for coverage. Cap cost, quarantine flakes, and never allow “skip evals” without a written exception.Tue, 21 Jul 2026 00:00:00 GMTEvals & production AI qualitycontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/embedded-pod-composition/https://www.brandlabs.app/blog/embedded-pod-composition/A Brandlabs embedded pod is a forward-deployed engineer, two senior builders, product design for AI trust UX, and compliance support when the domain demands it. It is not a slide team with a junior implementer. Roles exist to keep evals, UX, and production code moving on one cadence.Mon, 20 Jul 2026 00:00:00 GMTForward-deployed deliverycontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/us-healthcare-startup-ai-buy/https://www.brandlabs.app/blog/us-healthcare-startup-ai-buy/Buy an AI build partner on evidence: BAA-ready sequencing, eval gates, audit logging, clinical workflow humility, and a fixed delivery model with kill criteria. Dec reels and model-name dropping are not diligence. Brandlabs expects to be interviewed this hard — and interviews buyers the same way.Sun, 19 Jul 2026 00:00:00 GMTHealthcare & Indiacontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/discovery-sprint-two-weeks/https://www.brandlabs.app/blog/discovery-sprint-two-weeks/A Brandlabs Discovery Sprint is two weeks to pressure-test thesis, data, compliance posture, and eval targets — then make an honest go/no-go call. It is not a free mini-build. The artifact is a plan you could hand to another team and still respect.Sat, 18 Jul 2026 00:00:00 GMTForward-deployed deliverycontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/regulated-ai-buyer-guide-india/https://www.brandlabs.app/blog/regulated-ai-buyer-guide-india/Indian regulated buyers should diligence AI vendors on purpose limitation, consent UX, data residency, how knowledge stays current, and what happens when the model is wrong. Brandlabs expects these questions — and asks them of our own lab products like Correct.Fri, 17 Jul 2026 00:00:00 GMTHealthcare & Indiacontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/twelve-to-twenty-week-pmf/https://www.brandlabs.app/blog/twelve-to-twenty-week-pmf/Brandlabs runs zero-to-one AI in four phases over 12–20 weeks: Foundation (evals + compliance spine), build to alpha, production-harden, then PMF hunt. Each phase has exit gates. Skipping gates does not save time — it moves failure later.Wed, 15 Jul 2026 00:00:00 GMTForward-deployed deliverycontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/telehealth-trust-ux/https://www.brandlabs.app/blog/telehealth-trust-ux/Telehealth for women’s health fails when UX ignores stigma, shared devices, and language. Brandlabs designs trust as product: phone-first auth, vernacular copy, clear consent, separate roles for patients and clinicians, and flows that work at home — lessons from women’s health and specialty telehealth engagements.Tue, 14 Jul 2026 00:00:00 GMTHealthcare & Indiacontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/whatsapp-health-bots-safety/https://www.brandlabs.app/blog/whatsapp-health-bots-safety/WhatsApp health bots need a safety engine, not only a prompt: intent routing, multilingual refusals, escalation to humans, content filters, and eval fixtures for jailbreaks and medical overreach. Brandlabs shipped these patterns on a WhatsApp assistant for women’s SRH support across Hindi, Hinglish, Marathi, Telugu, and English.Sun, 12 Jul 2026 00:00:00 GMTHealthcare & Indiacontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/voice-ai-low-literacy-health/https://www.brandlabs.app/blog/voice-ai-low-literacy-health/Voice is the right interface when typing fails — but only if latency is near-instant and vernacular STT/TTS is solid. Brandlabs rebuilt a women’s health voice assistant on Pipecat with Sarvam for vernacular speech and cut multi-second delays that made phone health support unusable.Fri, 10 Jul 2026 00:00:00 GMTHealthcare & Indiacontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/india-msme-compliance-ai/https://www.brandlabs.app/blog/india-msme-compliance-ai/Indian MSME compliance is a visibility problem at impossible scale: tens of thousands of obligations shifting constantly. Correct, built in the Brandlabs lab, turns that chaos into dashboards, calendars, and AI-assisted checklists — with clear boundaries so the product helps operators without pretending to be a lawyer.Wed, 08 Jul 2026 00:00:00 GMTHealthcare & Indiacontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/evals-are-the-product/https://www.brandlabs.app/blog/evals-are-the-product/If you cannot measure whether an AI feature is good, you do not have a product — you have a demo. Brandlabs scopes eval suites (functional, qualitative, safety) as deliverables in week 1–3 so every later commit has a pass/fail bar.Sun, 12 Apr 2026 00:00:00 GMTEvals & production AI qualitycontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/shipping-on-fridays/https://www.brandlabs.app/blog/shipping-on-fridays/Friday demos work when they show evidence against eval targets and compliance gates — not polished slides. Brandlabs ships weekly inside client Slack with an explicit list of what is demoable, what is blocked, and what risk we are accepting.Sat, 28 Mar 2026 00:00:00 GMTForward-deployed deliverycontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/compliance-as-architecture/https://www.brandlabs.app/blog/compliance-as-architecture/HIPAA and DPDP fail late when they are treated as a launch checklist. Brandlabs designs access control, audit logging, consent, data minimization, and vendor BAAs into the first architecture so eval gates and product UX inherit the same posture.Wed, 04 Mar 2026 00:00:00 GMTRegulated AI architecturecontact@brandlabs.app (Yash Nerkar)https://www.brandlabs.app/blog/postmortems-as-regressions/https://www.brandlabs.app/blog/postmortems-as-regressions/AI incidents repeat when teams write prose postmortems and never encode the failure. Brandlabs requires each postmortem to produce at least one eval fixture and a CI gate so the same failure cannot silently return on the next model or prompt change.Sun, 15 Feb 2026 00:00:00 GMTEvals & production AI qualitycontact@brandlabs.app (Harsh Vitra)https://www.brandlabs.app/blog/forward-deployed-engineering/https://www.brandlabs.app/blog/forward-deployed-engineering/A forward-deployed engineer (FDE) at Brandlabs owns the engagement from inside your Slack: architecture, hardest implementation, eval discipline, and weekly evidence. They are not a PM layer and not a slide consultant — they ship.Fri, 30 Jan 2026 00:00:00 GMTForward-deployed deliverycontact@brandlabs.app (Harsh Vitra)