Iraq

I am a radiologist and I have about 85,000 radiology reports in .docx format (English), all stored in one main folder on my Windows laptop.I need a completely local, offline semantic search tool that helps me quickly find similar past reports for a new case. Main requirements: <ol><li>File renaming &amp; anonymisation</li><li> </li></ol><ul><li class="ql-indent-1">Scan one selected folder containing all my .docx reports.</li><li class="ql-indent-1">Automatically rename each file to a generic ID like RPT_00001.docx, RPT_00002.docx, etc.</li><li class="ql-indent-1">Create and save a local CSV file (e.g. mapping.csv) that maps: old_filename → new_ID.</li><li class="ql-indent-1">Provide a small script or button so I can run the renaming again later for new reports in the same folder.</li></ul><ol><li> </li><li>Replace old Arabic consultant name (mandatory)</li><li> </li></ol><ul><li class="ql-indent-1">At the bottom of every report there is an old Arabic consultant name and title (same pattern in all reports).</li><li class="ql-indent-1">I need a batch tool that replaces this old Arabic name + title with my new Arabic name and qualification in all existing reports and also for any new reports I add later.</li><li class="ql-indent-1">I will provide you with the exact old Arabic text and the new Arabic text.</li></ul><ol><li> </li><li>Semantic index using embeddings + vector database</li><li> </li></ol><ul><li class="ql-indent-1">Read the text of each report.</li><li class="ql-indent-1">Generate an embedding for each report (OpenAI embeddings or a suitable alternative – we can discuss).</li><li class="ql-indent-1">Store the vectors in a local vector database (FAISS or Chroma or similar) inside the project folder.</li><li class="ql-indent-1">No cloud storage and no external servers – everything must run on my Windows laptop only.</li></ul><ol><li> </li><li>Local search interface (Windows)</li><li> </li></ol><ul><li class="ql-indent-1">I prefer a simple solution: for example, a small local web app running on http://localhost:xxxx, started by double-clicking a .bat or .exe file.</li><li>The interface should include:</li><li class="ql-indent-1"> </li><li class="ql-indent-2">A text box where I can type or paste a new case description (findings / impression).</li><li class="ql-indent-2">A Search button.</li><li>A results list with similar old reports, showing for each result:</li><li class="ql-indent-2"> </li><li class="ql-indent-3">Rank/order.</li><li class="ql-indent-3">File ID / filename (e.g. RPT_00081.docx).</li><li class="ql-indent-3">A short snippet (first lines of the report).</li><li class="ql-indent-3">A similarity score (e.g. 87%, 82%, etc.).</li><li>A button/link to:</li><li class="ql-indent-3"> </li><li class="ql-indent-4">View full report text inside the app, and</li><li class="ql-indent-4">Open the original .docx file in Word.</li><li class="ql-indent-3"> </li><li class="ql-indent-2"> </li><li class="ql-indent-1"> </li></ul><ol><li> </li><li>Paged results in groups of 5</li><li> </li></ol><ul><li>When I search, the tool should:</li><li class="ql-indent-1"> </li><li class="ql-indent-2">First show the top 5 most similar reports (sorted from highest to lowest similarity).</li><li class="ql-indent-2">Have a “Show more” (or similar) button to show the next 5 results (6–10).</li><li class="ql-indent-2">If I click again, show the next 5 (11–15), and so on.</li><li class="ql-indent-2">All results (first 5, second 5, third 5, etc.) must always be ordered from highest to lowest similarity score.</li><li class="ql-indent-1"> </li></ul><ol><li> </li><li>Rebuild index when I add new reports</li><li> </li></ol><ul><li class="ql-indent-1">Provide a clear option/button like “Select folder &amp; rebuild index”.</li><li class="ql-indent-1">I choose the folder, and the tool rebuilds the semantic index to include the new reports.</li></ul><ol><li> </li><li>Portable folder</li><li> </li></ol><ul><li class="ql-indent-1">All code, data, index, and scripts should live inside one main folder so I can copy it to another Windows machine if needed.</li><li>A short step-by-step guide (PDF or DOC) explaining how to:</li><li class="ql-indent-1"> </li><li class="ql-indent-2">Set up any dependencies,</li><li class="ql-indent-2">Start the tool,</li><li class="ql-indent-2">Select the reports folder and rebuild the index,</li><li class="ql-indent-2">Use the search and open reports.</li><li class="ql-indent-1"> </li></ul><ol><li> </li><li>Data privacy</li><li> </li></ol><ul><li class="ql-indent-1">You will work only with dummy data. I will not send you real patient reports.</li><li class="ql-indent-1">At the end, you provide me with the full working folder and clear instructions, and I will run it locally on my real dataset.</li></ul><ol><li> </li></ol> Tech preferences (flexible): <ul><li>Python backend.</li><li>Vector DB: FAISS or Chroma (or your recommendation).</li><li>Simple local web UI (Flask/FastAPI + HTML/JS) or other lightweight desktop UI.</li></ul>

Local semantic search tool for 85,000 radiology reports Windows, offline

About the Client

Useful links

Company

Information

Content Writers

Freelancers

Freelance Jobs

Online Jobs

Designers

Developers

Translations