Agentic Engineer

A hands-on engineering role focused on building, fine-tuning, evaluating, and operating specialized AI agents for enterprise clients, end to end. The work blends applied research with production engineering — from choosing the right agent pattern to ensuring the resulting system is observable, cost-efficient, and reliable at scale.

• Own agent design from problem framing to production, selecting patterns such as tool-use orchestration, RAG, multi-step planning, and reflection loops
• Curate datasets and run LoRA or full supervised fine-tunes when off-the-shelf models fall short, evaluating them against a written harness
• Engineer prompt architecture, retrieval strategy, chunking, reranking, response caching, and eval-driven prompt iteration
• Optimize for cost and latency through quantization, batching, streaming, response budgets, and fallback strategies
• Instrument observability from day one, including tracing, regression suites, online evals, and drift alerts

• 4+ years of software engineering experience, with at least 1 year shipping LLM-based products to production
• Strong Python skills, plus enough Type Script or Go to ship glue and APIs
• Hands-on production experience with at least one major model provider (Anthropic, OpenAI, or open-weight Llama or Mistral on self-managed infrastructure)
• Practical experience with fine-tuning workflows — LoRA, full SFT, dataset curation, eval harnesses such as lm-eval-harness or equivalents, and basic reward modeling
• Production RAG experience covering vector stores, embedding choice, chunking strategies, reranking, and hybrid retrieval
• Cost and latency awareness, with the ability to read a billing dashboard or flamegraph and identify the right lever to pull
• Comfort with at least one major cloud (AWS, GCP, or Azure) and one container orchestrator (ECS, Kubernetes, or equivalent)

• Production experience with quantization techniques such as GPTQ, AWQ, or bitsandbytes
• Having built or contributed to an evaluation framework that runs in CI
• Experience working with on-prem or VPC-deployed models for compliance reasons