I’m building an in-house search experience that feels as fluid as ChatGPT yet stays completely private to our company data. The core of the project is an AI-powered engine able to understand natural language queries, crawl and index multiple internal sources, surface results with smart filters, and generate concise summaries and actionable insights for end-users. Tech stack is already defined: FastAPI will serve the endpoints, ElasticSearch will hold both keyword and vector indices, PyTorch will host any custom models, and OpenAI embeddings will provide the semantic layer. Everything must remain deployable on our own infrastructure. Content to index spans public and gated company websites, PDFs, financial and technical reports, department file shares, plus structured business profiles. Access control matters, so each dataset needs a private index with role-based permissions. The work includes: • Building the ingestion pipeline with incremental crawling, OCR for PDFs, and metadata enrichment • Training or fine-tuning a PyTorch model for vector search, hooked to ElasticSearch • Designing fluent NLP search endpoints (FastAPI) that return ranked results, smart filters, and summarised answers in one call • Implementing analytics that extract high-level insights from aggregated query logs and content trends • Supplying clean, readable code, Docker files, and setup documentation so our dev-ops team can reproduce the stack end-to-end Acceptance criteria will be a working demo on our staging server, able to answer conversational queries across all sources in under two seconds and passing an internal relevance/accuracy benchmark that we will share during kickoff. If this sounds like your wheelhouse, let’s set up a quick call and dive into the specifics.