Artificial Intelligence Engineer

📌AI Engineer

Our client, is looking for an AI Engineer to own and evolve a production conversational AI system. This assistant lets users ask natural-language questions and get back formatted answers, charts, and tables — powered by a LangGraph tool-calling agent, guardrails, and retrieval-augmented generation at scale.

What You'll Do

• 🔹Extend the core chat pipeline: guardrails, query reformulation, embedding-based tool routing, agent orchestration, and response formatting
• 🔹Maintain and grow a large domain tool layer over an internal API (schemas, auth checks, pagination, error handling)
• 🔹Support a live production system: triage incidents, investigate latency/quality regressions, improve telemetry and runbooks
• 🔹Improve RAG retrieval quality using PostgreSQL full-text search and pgvector, including hybrid approaches
• 🔹Build evaluation practices that gate model/prompt changes: adversarial datasets, tool-selection accuracy, shadow traffic, canary rollouts, auto-rollback
• 🔹Reduce LLM cost/latency via token telemetry, prompt trimming, caching, and model tiering
• 🔹Strengthen security: tenant-scoped data access, server-side authorization, and prompt-injection defenses
• 🔹Expand automated testing across commit, PR, nightly, and release gates

Must-Haves

• 🔹3+ years building production backend services, including at least one shipped LLM-backed feature
• 🔹Proven ability to take ownership of unfamiliar codebases safely using tests and traces
• 🔹Strong async Python with strict typing discipline (mypy)
• 🔹Hands-on experience with LangChain, LangGraph, or similar agent frameworks (tool/function calling)
• 🔹Solid PostgreSQL skills: schema design, performance tuning, migrations, connection pooling
• 🔹Experience operating live services: diagnosing issues from telemetry, rollback judgment
• 🔹Understanding of AI security boundaries — treating model/retrieved content as untrusted, enforcing authZ outside the model
• 🔹Comfort debugging across service boundaries using traces and correlation IDs
• 🔹Familiarity with retries, backoff, circuit breakers for rate-limited APIs
• 🔹Testing beyond unit tests — contract tests, eval of non-deterministic systems

Nice to Have

• 🔹Experience with vendor-to-in-house system handovers
• 🔹Azure (Container Apps, OpenAI deployments, Key Vault)
• 🔹Terraform/OpenTofu, CI/CD pipelines
• 🔹Vector search/RAG at scale, chunking, reranking
• 🔹LLM-as-judge evaluation awareness
• 🔹Guardrails frameworks (e.g., NeMo Guardrails)
• 🔹Modern React/Typ…