Overview

Join 2care.ai, a startup founded by experienced entrepreneurs backed by Silicon Valley and Indian health-tech angels, tackling the $430B global elder-care market with an AI Health Voice Agent, WhatsApp workflows, and a unified medical-data platform. We are building in India to ship to the world (starting with the USA). This role offers early ESOPs and a chance to make a significant impact in a mission-driven environment.

Responsibilities

Initially, you will:

• Ingest & clean at scale: Expand the Python pipeline for converting PDFs, DICOMs, and transcripts into de-identified FHIR resources stored in PlanetScale and Pinecone.
• Prompt engineering in production: Design, A/B-test, and version prompts for Symptom, Risk, and Treatment agents, iterating daily to reduce hallucinations and cost.
• Model fine-tuning: Distil or LoRA-tune open-weight LLMs and vision models when needed; deploy behind lightweight Python micro-APIs consumed by the TypeScript stack.
• Evaluation & observability: Build auto-evaluation harnesses for accuracy, bias, privacy leakage, and token spend; surface visually for clinicians.
• Privacy-first MLOps: Encrypt feature stores, log every model and prompt version, and enforce HIPAA/SOC 2 guardrails in CI.

Subsequently, you will:

• Prototype in Python (PyTorch, HuggingFace, LangChain) and deploy behind Next.js API routes via Triton or FastAPI, containerized for AWS GPU autoscaling.
• Maintain a vector store (Pinecone / Weaviate) and retrieval layer feeding agents end-to-end.
• Collaborate with full-stack engineers to expose models through our in-house AI-SDK and monitor them. ** Day-to-day toolset ** Python, PyTorch, Hugging Face, LangChain/LangGraph, Jupyter, CUDA, Pinecone, Redis, PlanetScale MySQL, Cloudflare Queues, AWS GPU autoscaling, Next.js API routes, in-house AI-SDK, OpenTelemetry, Sentry. Current stack: NextJs(TypeScript) • MySql (Planetscale) • Cloudflare(R2,Queues, Workers) • Redis • AI Integration with Claude, OpenAI and Gemini• Whatsapp Business Integration • Vercel • Github with actions for CI/CD

Requirements

• Minimum 3 years of experience.
• Have shipped at least one NLP, CV, or LLM feature to production and can explain how success was measured.
• Write solid Python and are comfortable tweaking TypeScript glue when needed.
• Know how to tune temperature, top-p, logit bias, and token windows.
• Either understand HL7/FHIR, ICD-10, RxNorm already or can learn them fast.
• Crave autonomy, move in one-week sprints, and enjoy talking directly to doctors and families about what works. *** Healthcare Tech experience a BIG plus. **

Six-month north-star

• 90 % of incoming records parse end-to-end with < 2 % field-level error.

• Hallucination rate across agents drops below 2 % while per-conversation cost falls 40 %.

• Every model call and prompt is traceable and audited, letting us onboard our first US provider without re-architecture.

• Sound like your kind of playground? Email your résumé, one LLM/ML project link, and a short note on why 2care.ai