Software Development Engineer III

What is Hunch?

• One Drop a Day: A small, hand-picked set of photo-verified profiles arrives daily - chosen for taste, lifestyle, and who's actually free to meet. So every profile gets real attention instead of a mindless swipe.
• Interest Becomes a Plan: When both people want to meet, Hunch's AI matchmaker turns mutual interest into a confirmed date with a real time and place - no logistics to manage.
• Every Match, Verified: Every profile is matched to a live photo before it appears in anyone's drop. No bots, no catfishing - just real, trustworthy connections.
• Built to Move: Chats close after seven days and users hold a few conversations at a time - nudges designed to get people out the door, not scrolling forever.

• Own backend features end-to-end — product problem to API, schema, migration, tests, and deployment. You solve the problem, not just the ticket.
• Build and run services across both stacks: TypeScript (Bun/Node, Prisma, Next.js) and Python (FastAPI, Pydantic, SQLAlchemy).
• Own async & event-driven work — queues, streaming ingestion, scheduled jobs, retries, and idempotency. (A lot of our load never touches a request handler.)
• Ship LLM-backed product features — tool calling, structured outputs, grounding against real data, and the failure handling that keeps them sane in front of users.
• Keep production legible — tracing, metrics, and alerts that fire only when action is needed. You own your deploys and their consequences.

• 5+ years building and running backend systems in production. You've been on the hook when things broke.
• Dual-stack proficiency: TypeScript/Node + Python — strong in one, genuinely productive in the other. This isn't a "Python nice-to-have" role.
• Relational databases at depth: Postgres or MySQL — schema design, migrations, indexing, and reading query plans to fix the slow ones.
• Streaming/event infrastructure in production: Kinesis or Kafka — producers, consumers, partitioning, replay, and handling backpressure and duplicates.
• Service design & delivery — clean contracts, sensible boundaries, and knowing when not to split a service. Hands-on with AWS, Docker, and CI/CD.
• End-to-end ownership — you've carried features from idea to production and lived with the results. Founding or product-engineer experience is a huge plus.
• Engineering hygiene — writing tests with the change (not after), disciplined releases, and caring whether the metric actually moved.

• ML Platform: Serving models behind APIs, vector retrieval (pgvector/FAISS), feature pipelines, and model versioning/rollout. Keeping models alive in production counts double.
• LLM Application Engineering: Agent loops, tool calling, structured output, and evaluating output quality programmatically. Self-hosted inference (vLLM, GPU serving, cost tuning) is a strong plus.
• Data at Scale: Redshift/BigQuery/Snowflake, and observability tools (OpenTelemetry, Datadog).
• Experimentation: A/B and shadow tests, guardrail metrics, and the discipline to read a flat result honestly.
• Domain Experience: Consumer social or dating products.

What We Offer

• Competitive pay along with an annual performance-linked incentive (PLI).
• Employee benefits like Medical Insurance.
• Participation in Employee Phantom Grants Program (Stock Options).

Life at Hunch

• Authenticity - Real before polished. We build for real people on real first dates, and we hire the same way. No theater. If something's not working, say so; if it is, ship it.
• Boldness - Dating got stuck because nobody wanted to break the pattern. We make the call on what to ship, what to cut, and what doesn't belong in the product, even when it's the harder choice.
• Curiosity - The best people here ask a lot of small, useful questions — about users, about the data, about each other's work. We trust people who keep learning over people who already know.

• Ownership, not theater - You own outcomes, not status updates. Fewer meetings, fewer dashboards, more shipping.
• Flexibility that's real - We care about the work, not the hours you sit at a desk. Take the afternoon, take the week - show up for the stuff that matters.
• Move fast, don't burn out - Hunch ships quickly because we cut the slow stuff, not because we work people into the ground. Velocity comes from focus, not from hours.
• Small team, big problems - You'll work on things that matter to millions of people trying to actually meet someone. Every role moves the product.