MicroHeal (formerly WinHealth )
Website:
microheal.in
Job details:
Senior AI Full-Stack Engineer/ Founding Engineer — SaaS/PaaS Platform + Agentic Coding
Company: Microheal — digital therapeutics (DTx) platform · Location: India (onsite) · Type: Full-time
The product:
Microheal runs care programs (weight management, gut health, CKD, diabetes, women’s health) for clinics and multi-speciality hospitals across India. Patients interact through a fleet of WhatsApp AI agents — onboarding, appointments, symptom logging, nutrition tracking, assessments, reminders — while care teams work from a multi-tenant dashboard. Three branded bot lines run on one codebase today; more clinic partners are onboarding, and we’re rebuilding the platform so that a new clinic goes live in hours, not days — spec-driven provisioning, automated QA simulation, compliance baked into the schema.
We’re a deliberately small team building platform, not features. You’d be engineer #1 on the product side.
The role
You own everything patient-facing and tenant-facing: the Python/LangGraph agent runtime, the Next.js care-team dashboard, the data-access layer on Cloud SQL Postgres (RLS-enforced tenant isolation), and the agent quality pipeline (golden conversation corpus, QA simulation, safety gates). This is a SaaS/PaaS engineering role at its core — multi-tenant isolation, config-driven behavior, tenant onboarding as a product surface — with AI agents as the workload running on that platform.
The second defining feature of the role: you will build with agentic coding agents, not just alongside them. Our engineering model is a minimal human team where coding agents do bulk implementation from a written specification suite, and humans own direction, review, and judgment. We need someone who has already worked this way in production — not someone curious to try it.
What you’ll own
• Agent runtime (Python/LangGraph): three bot lines on one graph — intent routing, a hard safety gate (emergency / medical-advice / pricing blocks enforced in code, not prompts), workflow agents for booking, onboarding, logging, reminders
• Multi-tenant platform (TypeScript/Next.js + Postgres): the care-team dashboard, tenant isolation via Row Level Security with per-turn tenant context, spec-driven per-clinic configuration (170+ settings generated, never hand-edited)
• Onboarding pipeline: the generators that turn a Clinic Spec (YAML) into provisioned tenants, Meta assets, runtime config, and dashboard views
• Quality as a system: golden conversation corpus + synthetic probes replayed by a QA simulation agent before any clinic goes live; you curate the corpus and own the release gate
• Review of agent-produced code: the coding agent drafts; you approve. Your review bar is the product’s safety bar
• On-call (primary, weekly rotation with our Platform Engineer): platform uptime; clinical escalations route to humans by design, so this is sane on-call, not 3 AM medicine
Must have:
1. 2 to 5 plus years building production SaaS/PaaS systems — multi-tenant products where tenant isolation, per-tenant configuration, and safe tenant onboarding were your problem, not someone else’s. You can point at the isolation mechanism you shipped and explain its failure modes
2. Demonstrated agentic coding experience — this is a hard requirement. You’ve used AI coding agents (Claude Code, Cursor, Copilot Workspace, Devin, or similar) as a daily driver on production code for months, not weeks.
Concretely, you can show us:
– how you structure specs/context so agents produce mergeable work
– your review workflow for agent-generated PRs — what you check first, what you never trust
– a time an agent got something subtly wrong and how your process caught it
3. Full-stack fluency: strong Python (async services, preferably FastAPI) AND strong TypeScript/React (Next.js). You don’t “specialize in one and dabble in the other” at this seniority
4. Postgres depth: schema design for multi-tenant data; RLS experience is a major plus, solid indexing/transaction fundamentals are the floor
5. LLM systems in production: agent loops, tool calling, evaluation harnesses, failure-mode analysis. Not demos — systems with real users and real consequences
6. Judgment for regulated domains: you instinctively ask “what must this system never do?” Healthcare experience is not required; healthcare-grade caution is
Strongly preferred :
• WhatsApp Cloud API / Meta Flows (templates, approval cycles, webhook signatures)
• LangGraph or comparable agent-orchestration frameworks
• GCP (Cloud Run, Cloud SQL, Secret Manager) — our infrastructure home
• Indic-language product experience (Hinglish, code-mixed input)
• Exposure to DPDP Act, HIPAA, or SOC 2 environments
• Test/eval culture: you’ve built simulation or replay harnesses for non-deterministic systems
Your first 90 days:
• Weeks 1–2: ship the fleet-critical security fixes with our Platform Engineer (webhook authentication, token hygiene, RBAC gaps — already spec’d, you execute)
• Weeks 3–6: land Phase 1 of the onboarding pipeline — spec validator + tenant provisioner, using the coding agent for bulk implementation and owning every review
• Weeks 7–12: stand up the QA simulation gate against our 33-conversation golden corpus; first new clinic goes live through the pipeline with you holding Gate 2
• Throughout: become the human whose judgment the whole spec suite is designed to amplify
What we are NOT looking for :
• ML researchers — our models come from APIs; our intelligence is orchestration, evaluation, and guardrails
• “AI enthusiasts” without production SaaS scars — the tenant isolation matters more than the prompting
• Prompt engineers without full-stack depth — this role ships services, schemas, and dashboards, not prompts
Why this role is unusual
• Real ownership, tiny team: you are the product engineering function. Every patient-facing decision has your name on it
• A genuinely spec-driven workflow: the platform’s architecture, schemas, trap lists, and work packages are written down and versioned — the coding agent builds from them, and so do you
• Healthcare stakes without healthcare bureaucracy: the safety gate, consent model, and audit trail are engineered into the platform, and you get to build that, not just comply with it
• The 2-human + agents experiment, for real: if you’ve wanted to prove what a minimal, agent-amplified team can ship, this is that job
Interview process :
- Intro call (30 min) — your SaaS/PaaS and agentic workflow story
- Agentic review exercise (90 min, async): review an agent-generated PR against a written spec — we evaluate what you catch, not whether you catch everything
- System design (60 min): multi-tenant onboarding pipeline under healthcare constraints
- Founder conversation: judgment, ownership, on-call expectations
To apply: reply through the channel where you found this post or email at shashank@microheal.in with your GitHub/LinkedIn and 3–4 lines on a system you shipped where a coding agent did a meaningful share of the implementation — and what you caught in review.
Compensation : 10-35 lacs ( cash + esops )
Click on Apply to know more.