📚 RAG Knowledge Base API

A FastAPI-based Retrieval-Augmented Generation (RAG) system that lets you upload documents, store them in GCP and MongoDB, and ask questions using Google Gemini AI.

---

🧱 Architecture

User → FastAPI → GCP Storage (file stored)
                → Gemini File API (content indexed)
                → Gemini File Search Store (RAG grounding)
                → MongoDB Atlas (metadata saved: id, gcp_link, filename)

User → /kb-ask or /upload-and-ask → Gemini AI → Answer

---

🚀 Running the Server

# Activate virtual environment
source venv/bin/activate

# Start server (auto-reloads on file changes)
uvicorn main:app --reload --port 8000

# Swagger UI available at:
# http://localhost:8000/docs

---

📬 Postman Collection Setup

1. Open Postman → New Collection → Name it RAG Agent
2. Set a Collection Variable: base_url = http://localhost:8000
3. Add the 5 requests below

---

📌 Endpoints

---

1. POST /upload

Upload a document to the knowledge base.

Uploads to GCP Storage, indexes in Gemini, and saves metadata in MongoDB.

Setting	Value
Method	POST
URL	{{base_url}}/upload
Body	form-data

Form fields:

Key	Type	Required	Description
file	File	✅	Select your PDF or TXT file
display_name	Text	❌	A friendly name (defaults to filename)

Response:

{
  "status": "success",
  "original_filename": "Todays_discussion.pdf",
  "gcp_link": "https://storage.googleapis.com/...",
  "id": "699dc9606f0d05f522ccaac2"
}

💾 Save the id and gcp_link — you'll need them for /kb-ask.

---

2. GET /list

List all uploaded documents with their IDs and GCP links.

Setting	Value
Method	GET
URL	{{base_url}}/list
Body	None

Response:

{
  "files": [
    {
      "id": "699dc9606f0d05f522ccaac2",
      "original_filename": "Todays_discussion.pdf",
      "display_name": "Todays_discussion.pdf",
      "gcp_link": "https://storage.googleapis.com/...",
      "uploaded_at_human": "2026-02-24 15:53:04 UTC"
    }
  ]
}

---

3. POST /upload-and-ask

Upload a document and immediately ask a question about it.

Does everything in one shot: upload → index → query → answer.

Setting	Value
Method	POST
URL	{{base_url}}/upload-and-ask
Body	form-data

Form fields:

Key	Type	Required	Description
question	Text	✅	Your question about the document
file	File	✅	Select your PDF or TXT file

Response:

{
  "question": "What is this document about?",
  "answer": "This document is about...",
  "selected_documents_count": 1
}

---

4. POST /kb-ask

Ask a question using documents already in the knowledge base.

Setting	Value
Method	POST
URL	{{base_url}}/kb-ask
Body	form-data

Form fields:

Key	Type	Required	Description
question	Text	✅	Your question
file_ids	Text	❌	Comma-separated MongoDB IDs from /list. Use all to search everything.
file_urls	Text	❌	Comma-separated GCP links from /list

4 modes of use:

Mode A — Default (no filter): uses your most recently uploaded document

question  →  What is this document about?

(leave file_ids and file_urls empty — automatically picks the latest upload)

Mode B — Search ALL documents:

question  →  Summarize everything
file_ids  →  all

Mode C — Query specific docs by ID:

question  →  What is the project codename?
file_ids  →  699dc9606f0d05f522ccaac2

(copy the id field from /list response)

Mode D — Query specific docs by GCP URL:

question   →  Who wrote this?
file_urls  →  https://storage.googleapis.com/rag-agent-docs-aryan/kb/...

ℹ️ For multiple documents in Mode C or D: comma-separate values. file_ids → id1,id2,id3

---

5. DELETE /delete/{id}

Delete a document from the knowledge base.

Removes from MongoDB, and best-effort cleanup from GCP and Gemini.

Setting	Value
Method	DELETE
URL	{{base_url}}/delete/699dc9606f0d05f522ccaac2
Body	None

Put the document id (from /list) directly in the URL.

Response:

{
  "status": "deleted",
  "id": "699dc9606f0d05f522ccaac2"
}

---

🔄 Typical Workflow

1. POST /upload          → upload your PDF
2. GET /list             → see all docs + copy the id
3. POST /kb-ask          → just send a question — auto uses your latest upload!
4. POST /kb-ask          → pass file_ids for a specific doc, or file_ids=all for everything
5. DELETE /delete/{id}   → clean up when done

Or the fast route:

1. POST /upload-and-ask  → upload + ask in one step
2. POST /kb-ask          → ask follow-up questions (auto picks your latest upload)

---

⚠️ Important Notes

Note	Detail
Default behaviour	/kb-ask with no file_ids or file_urls automatically uses your most recently uploaded document.
Search all docs	Pass file_ids=all to search across your entire knowledge base.
File expiry	Gemini File API files expire after 48 hours. Re-upload if /kb-ask with file_ids stops working on old docs.
Legacy docs	Docs uploaded before this version show gcp_link: null in /list — re-upload them to enable per-document querying.
Supported formats	PDF, TXT, MD, DOCX, PNG, JPG (any file Gemini can parse)

---

🔐 Environment Variables (.env)

GEMINI_API_KEY=...
MONGO_URI=mongodb+srv://...
GCP_PROJECT=your-gcp-project
GCP_BUCKET_NAME=your-bucket-name
FILE_SEARCH_STORE_NAME=my-knowledge-base-store
MODEL_ID=gemini-2.0-flash