yuyu1623/bid_agent

This is a desktop tool designed to help people who respond to government or business tenders. When an organization puts out a tender, they publish a formal document describing what they need and how bids will be evaluated. Reading and extracting the key details from those documents is time-consuming, especially when they come as PDFs, Word files, or scanned images. This tool automates that work. The project has two main parts: a Python backend that handles all the document processing and AI analysis, and a desktop application (built with Electron) that gives users a graphical interface. Users upload a tender document through the desktop app, and the backend figures out the best way to read it, whether that means parsing a PDF directly, using an external document intelligence service called MinerU, or extracting text from images using OCR tools. Once the document is read and converted to structured text, the tool sends it to an AI language model for analysis. Four separate analysis tasks run at the same time, each focusing on a different section of the tender: the bidder instructions and project overview, the technical requirements, the qualification and eligibility criteria, and the scoring rules. Results appear in the desktop app organized under those five categories, and users can edit the parsed content and re-run the analysis if something looks wrong. The tool supports both streaming output, where results appear as the AI produces them, and regular output that waits until analysis is complete. It works with any AI service that follows the OpenAI API format, so users can point it at different models by changing a configuration file. MinerU can also run locally rather than through a cloud API if the user has it installed. The project is described as actively in development, with planned additions including Word and PDF export, template-based bid document generation, and automated checks for disqualifying conditions.