Build and operate ML-backed backend systems (RAG, retrieval, ranking, embeddings) powering large-scale consumer experiences. This is a hands-on role with real ownership across quality, latency, and cost.

A hands-on backend engineering role building the core AI/ML-backed systems that power consumer experiences at scale. This role is not learn ML on the job. You must already be effective in building and shipping ML-adjacent backend systems (RAG/retrieval, embeddings, ranking, evaluation hooks, feature pipelines) and making pragmatic tradeoffs across quality, latency, and cost. We cannot upskill ML fundamentals in this engagement; candidates must demonstrate prior delivery of ML- backed backend systems in production.

The Engineer will :

- Build and operate core backend services for AI product runtime: orchestration, state/session, policy enforcement, tools/services integration

- Implement retrieval + memory primitives end-to-end: chunking, embeddings generation, indexing, vector search, re-ranking, caching, freshness and deletion semantics

- Productionize ML workflows and interfaces: feature/metadata services, online/offline parity, model integration contracts, and evaluation instrumentation

- Drive performance and cost optimization (P50/P95 latency, throughput, cache hit rates, token/call cost, infra efficiency) with strong SLO ownership

- Add observability-by-default: tracing, structured logs, metrics, guardrail signals, failure taxonomy, and reliable fallback paths

- Collaborate with applied ML on model routing, prompt/tool schemas, evaluation datasets, and release safety gates

What were looking for (must-have):

- 6 - 10 years building backend systems in production, including at least 2 - 3 years on ML/AI-backed products (search, recommendations, ranking, RAG, or assistants)

- Practical ML chops : able to reason about embeddings, vector similarity, reranking, retrieval quality, evaluation metrics (precision/recall, nDCG, MRR), and data drift - without needing training

- Experience implementing or operating RAG pipelines (document ingestion, chunking strategies, indexing, query understanding, hybrid retrieval, re-rankers)

- Strong distributed systems fundamentals: API design, idempotency, concurrency, rate limiting, retries, circuit breakers, and multi-tenant reliability

- Comfort with common ML/AI platform components: feature stores/metadata, streaming/batch pipelines, offline evaluation jobs, A/B measurement hooks

- Ability to ship end-to-end independently: design ? build ? deploy ? operate in a fast-moving environment

Bonus (nice to have) :

- Agentic runtime / tool-calling patterns, function calling schemas, structured outputs, safety/guardrails in production

- Prior work with FAISS / Milvus / Pinecone / Elasticsearch hybrid retrieval, and model serving stacks

- Kubernetes + observability stack depth (OpenTelemetry, Prometheus/Grafana, distributed tracing), plus privacy controls for user data

Screening expectation (signals we will check) :

- Can explain a shipped ML-backed system end-to-end and defend key tradeoffs (quality vs latency vs cost)

- Can discuss retrieval evaluation and failure modes (bad chunking, stale index, poor recall, hallucinations, noisy sources) and how they mitigated them

- Can write production-grade code and operationalize it (tests, metrics, dashboards, runbooks)