Semantic Search Tool for Radiology Reports
Budget: $250 – $750 USD
I am a radiologist and I have about 84,000 radiology reports (~1.2 GB).
Each report file contains only the exam name (e.g. "CT chest") and the radiological findings. There are no patient names, IDs, age, or history inside the text.
However, the file names themselves currently contain patient names in English (for example: "Ali Mohammad has a.docx"). For privacy reasons, I need these file names to be automatically renamed to generic IDs (e.g. RPT_00001.docx, RPT_00002.docx, …) on my local machine before building the search index.
I want a simple LOCAL tool on my Windows laptop that can:
1. Read and index all my reports using semantic search (embeddings).
2. When I type a new case description in free text (modality, body part, key findings, etc.), the tool returns the top 5–10 most similar old reports based on MEANING, not just keywords. Different wording should still match similar cases.
3. Use Python + OpenAI embeddings (or similar) + a vector database (FAISS, Chroma, etc.).
4. Run completely on my laptop with a simple interface (desktop app or small web app running on localhost) where I can paste my query and see the closest reports with their IDs and full text.
5. Be portable: all data + index + code in one folder so I can move it to another laptop.
Important notes:
- You do NOT need direct access to my real reports. You can develop the tool using dummy data, and I will run it on my own files.
- My English is not perfect, so I prefer clear and simple communication.
Please:
- Briefly describe how you would implement this (including how you will handle renaming the files and building the semantic search index).
- Share similar projects you have done (semantic search, document search, OpenAI embeddings, etc.).
- Give me an estimated fixed price and timeframe.
I specifically want AI-based semantic search (embeddings), not simple keyword search.
Each report file contains only the exam name (e.g. "CT chest") and the radiological findings. There are no patient names, IDs, age, or history inside the text.
However, the file names themselves currently contain patient names in English (for example: "Ali Mohammad has a.docx"). For privacy reasons, I need these file names to be automatically renamed to generic IDs (e.g. RPT_00001.docx, RPT_00002.docx, …) on my local machine before building the search index.
I want a simple LOCAL tool on my Windows laptop that can:
1. Read and index all my reports using semantic search (embeddings).
2. When I type a new case description in free text (modality, body part, key findings, etc.), the tool returns the top 5–10 most similar old reports based on MEANING, not just keywords. Different wording should still match similar cases.
3. Use Python + OpenAI embeddings (or similar) + a vector database (FAISS, Chroma, etc.).
4. Run completely on my laptop with a simple interface (desktop app or small web app running on localhost) where I can paste my query and see the closest reports with their IDs and full text.
5. Be portable: all data + index + code in one folder so I can move it to another laptop.
Important notes:
- You do NOT need direct access to my real reports. You can develop the tool using dummy data, and I will run it on my own files.
- My English is not perfect, so I prefer clear and simple communication.
Please:
- Briefly describe how you would implement this (including how you will handle renaming the files and building the semantic search index).
- Share similar projects you have done (semantic search, document search, OpenAI embeddings, etc.).
- Give me an estimated fixed price and timeframe.
I specifically want AI-based semantic search (embeddings), not simple keyword search.