Description

I am a radiologist and I have about 85,000 radiology reports saved as .docx files in one main folder on my Windows laptop.
I need a fully local search tool that works by meaning (semantic search), not simple keywords.

What I need the tool to do:
1. Read all .docx reports from one selected folder.
2. Automatically rename the reports to anonymous IDs (e.g. RPT_000001.docx) and create a local CSV mapping file (old name → new ID).
3. Build a semantic index using text embeddings + a local vector database (FAISS, Chroma, or similar).
4. Provide a simple local interface (desktop app or small local web app in the browser) with:
• One text box to paste or type the new case description (modality, body part, key findings).
• A Search button.
• Results list sorted from most similar to least similar.
• For each result: rank, file ID/name, short snippet, and a similarity score in % (for example 87%, 82%).
• Ability to open the full .docx report with one click in Microsoft Word.
5. Results pagination by 5 reports:
• First show the top 5 most similar reports.
• A “Show more” (or similar) button to show the next 5 (6–10), ordered by similarity.
• Clicking again shows the next 5 (11–15), and so on.
• All results must always be ordered from highest to lowest similarity.
6. I already use a Word macro: when I choose a report, I click the macro and it:
• Saves the report into a monthly folder,
• Renames the file according to my system,
• Prints the report.
The new search tool must not break this workflow. When I open a report from the search results, it should open as a normal Word document so I can run my existing macro as usual. If you can integrate a simple button or shortcut that triggers the macro, that is a plus.
7. The tool must be fully local:
• No cloud, no external API calls with my real reports.
• You can work with dummy data and send me the scripts; I will run them on my real reports myself.
8. The whole solution should be in one portable folder (code + data + index + simple instructions) so I can copy it to another Windows laptop.
9. Add a “Rebuild index” function so I can re-index if I add new reports in the future.
10. Mandatory requirement:
A batch module to replace the old Arabic consultant name and title at the bottom of all reports with my new Arabic name and qualification. This should work automatically on all reports.

Technical hints (not strict, you can propose better):
• Python backend.
• FAISS or Chroma (or similar vector DB).
• Any reliable embeddings (OpenAI, local model, etc.), but final tool must run fully local on my side.

My English is not perfect, so please write in simple clear English when we chat.
Please send:
• A short explanation of your approach.
• Example of similar projects (document search / semantic search).
• Estimated time and milestones.

Budget around 475 USD for the complete solution.