Job Description & Details
This is a serious enterprise-grade AI gig where they aren't looking for résumé fluff—they explicitly demand authentic experience with modern generative stacks and zero document tampering. You will be building real production-ready agentic systems and LLM workflows rather than just playing around in Jupyter notebooks. If you know your way around transformers, RAG architectures, and actual cloud deployments, this is a solid opportunity to ship impactful AI features.
What You'll Actually Be Doing
Expect to spend your days knee-deep in architecting Retrieval-Augmented Generation (RAG) pipelines, setting up robust evaluation loops, and orchestrating multi-agent frameworks using tools like LangGraph or LlamaIndex. You will be bridging the gap between raw machine learning models and scalable production systems, which means dealing with messy unstructured data, optimizing token usage, and figuring out how to stop LLMs from hallucinating in production.
The Core Tech Stack
Python is your primary weapon here, and you'll need to write production-grade code that goes beyond basic scripting—think Docker containerization, clean CI/CD pipelines, and solid SQL database skills. On the AI side, absolute fluency with PyTorch or TensorFlow, transformer architectures, and API integrations with OpenAI, Claude, or Azure OpenAI is non-negotiable. You'll also need hands-on experience with vector databases and observability platforms like Langfuse or Arize to track model drift and costs.
Interview Expectations
Be ready to defend your architectural decisions when designing a complex RAG system under tight latency constraints, particularly around chunking strategies, re-ranking, and context window optimization. The hiring manager is going to probe deep into your understanding of agentic workflows and tool-calling failures, specifically looking to see if you understand the edge cases where multi-agent systems break down in production.
Application Advice
Make sure your resume clearly highlights production deployments of LLMs rather than just academic or toy projects. Explicitly feature keywords like LangChain, PyTorch, vector databases, and MLOps tooling so you easily clear the strict ATS filters, and ensure your timeline matches your actual depth with these modern AI stacks since they have zero tolerance for discrepancies.