About Zumlo
Zumlo is an always-on well-being companion—one place for immediate help, gentle structure, and progress you can see. We unify mind, body, emotions, and relationships through timely support, a caring community, and personalized guidance that fits real life.
What we stand for : Human first
complex
Privacy & trustEvidence over hypeInclusive by defaultThe role (why it’s rare)
AI isn’t a bolt-on here—it’s the nervous system of the product. We’re hiring a senior who has already shipped LLM + retrieval to production and wants end-to-end ownership : problem framing, modeling / orchestration, evaluation, privacy / safety, and the Python services that make it real.
What you will own
Product AI (end-to-end)
Build AI across product surfaces : conversational help, guided steps, tailored activities, “what to do next,” summaries / explanations, and safety checks—grounded with retrieval and citations.Turn fuzzy needs into robust flows : prompt design, tool / function calling, JSON-schema outputs, fallbacks, streaming, controllable tone and safety boundaries.Retrieval & knowledge
Do RAG right : chunking / segmenters, embeddings, vector DBs (FAISS / qdrant / Pinecone / Milvus), hybrid semantic + re-rankers (BGE / ColBERT), dedupe, freshness policies, provenance.Rapid test & evaluation loop
Make eval routine : golden sets, adversarial suites, shadow evals, canaries, online metrics tied to user outcomes. Capture in-app feedback and close the loop weekly.Safety, privacy, governance
PHI / PII redaction, prompt-injection defenses, output guardrails, rate limits, audit trails, safe logging. Clear data-handling notes for the team.Backend & reliability
Own Python services : FastAPI, Postgres (schemas / migrations), Redis, task queues, retries / idempotency, auth / RBAC, feature flags.Observability first : logs / metrics / traces, alerting that matters, simple SLOs—systems that are predictable and calm.Data & experimentation
Trustworthy event tracking, simple SQL cohorting, per-feature cost / latency dashboards, A / B hooks so product / growth can run honest tests.Exploration & de-risking
Evaluate models / embeddings, inference servers (vLLM / TensorRT-LLM), compression / quantization, token-efficiency. Prove value with small, cheap spikes before big changes.Collaboration & leadership
Partner with product, mobile (React Native / TS), and platform.Review PRs, write concise docs / runbooks, mentor juniors, and help hire the next 1–2 great engineers.Must-have experience
7+ years software engineering with 4+ years Python shipping production APIs / services.Production LLM + RAG you can discuss end-to-end : retrieval, orchestration, evaluation, and user impact.RAG depth : embeddings, vector DBs (FAISS / qdrant / Pinecone / Milvus), hybrid search, re-rankers, citation strategies.Backend foundation : FastAPI / Django / Flask, Postgres / SQL, Redis, queues (Celery / RQ), testing (pytest), CI / CD, containers.Eval mindset : offline metrics + online behavior; sample-size sense; knows when to ship, iterate, or kill.Security / privacy : least-privilege, secrets, encryption, safe logging; comfortable with sensitive data.Strong plus
Well-being / health context; HIPAA-aware practices.Azure & Azure DevOps pipelines; GPU inference; streaming responses.Telemetry for AI : prompt / version tracking, per-feature cost / latency, drift monitors.Worked across US + India user bases / time zones.How you work (what we value)
Builder energy : ship → measure → iterate.Creative + logical : playful with ideas, strict with evidence.Product-curious : start with the user problem and “definition of good.”Kind, direct, low-ego : crisp commits / PRs, generous feedback.Owner’s mindset : reliable, documented, observable.Work setup
Remote-first in India, collaborating closely with a small core team.Preference for Ahmedabad for an eventual in-person cadence; open across India for the right fit.