lijingjim/j-aeis

J-AEIS is an enterprise-style AI assistant built with the OpenAI SDK, Streamlit, and function calling. You upload a file or describe what you need, and it analyzes the data and produces a report, running as a web app in your browser rather than a hosted service. The assistant bundles several built-in skills. It can analyze spreadsheet data using a structured what, why, so-what framing, review contracts with a two-layer check that combines keyword scanning with a deeper language model reading for things like unequal terms or clauses buried in attachments, and parse general files including PDFs, Word documents, PowerPoint, images, HTML, and ZIP archives. It also builds a searchable knowledge base from uploaded documents using vector search through ChromaDB, so it can look up and cite specific passages. Reports are generated by first planning an outline, then expanding each section, then formatting the result, and exporting to both Word and PDF. On the technical side, the agent can create new tools on the fly inside a sandbox when its built-in tools are not enough, streams its output token by token so you see a long report as it is written rather than waiting for the whole thing, and runs multiple tool calls in parallel using a thread pool to cut down response time. It automatically retries on timeouts or rate limits, and a twelve point self-check pass reviews its own draft reports for issues like unsupported claims or inconsistent formatting before showing them to you. Conversations are saved to a local SQLite database so they survive a page refresh. To run it, you install Python dependencies with a provided batch script or pip, then launch the app with another batch script and open it in your browser, entering your API key in the sidebar on first use. Configuration such as the API key, base URL, and model name is set through a config file or environment variables. This is aimed at business or technical users who want a local, customizable AI assistant for document review, spreadsheet analysis, and report writing, rather than end consumers.