RAG AI Engineer

Job Overview:

You’ll work on Invictus AI Insights, our document-intelligence platform: documents come in, get extracted, enriched, chunked, and vectorized, then served back through a retrieval-and-generation (RAG) layer. It’s a Python codebase running on Azure. You’ll own real features and fixes in production with close mentorship — pitched for someone early in their career who wants to go deep on applied AI and cloud backends.

Job Responsibilities:

• The document pipeline (extraction → enrichment → chunking/vectorization → indexing) and the RAG gateway that answers questions over those documents.
• Writing and testing Python services, reproducing issues, and shipping well-scoped improvements.
• Making the pipeline more accurate, faster, and more resilient.

Requirements:

• 1–2 years of backend engineering or strong internship/project experience.
• Solid Python, including a working understanding of async/await (asyncio) and the basic difference between concurrency, threading, and async.
• Hands-on experience with at least one cloud provider (Azure, AWS, or GCP) and a basic grasp of how services talk to each other — with a willingness to go deep on Azure fast.
• Basic working knowledge of LLMs — what tokens, context windows, and embeddings are and what RAG is at a high level.
• Good debugging instincts — you reach for logs and reproduction steps before guessing, and you don’t give up when something’s unclear.
• Comfortable with Git, REST/HTTP, and JSON.
• Fluency with AI-assisted dev tools (Claude Code, Cursor, or similar).
• A fast learner who likes to go deep — we care more about how quickly you pick things up than how much you already know.

Nice To Have:

• Hands-on Azure specifically — Function Apps, Service Bus, Cosmos DB, Blob Storage.
• Azure OpenAI / GPT-4o, or document-AI/OCR (Azure Document Intelligence).
• RAG, vector search, or Azure AI Search experience.
• Pydantic or FastAPI.
• Testing with pytest; CI/CD with GitHub Actions.
• Awareness of observability tooling (Application Insights, OpenTelemetry).