A full-stack, open-source web application for semantic case management using TiDB Serverless (vector search), FastAPI, and React. It uses Amazon Bedrock Titan-V2 for text embeddings and Claude 3 Sonnet for LLM-based summarization in semantic search.

Access the live Semantic Case Manager app here

• Amazon Bedrock Titan-V2 for text embeddings

• TiDB for Emeddings storage

• TiDB Serverless vector search for semantic retrieval and

• Claude 3 Sonnet for LLM-based summarization

• Multi-step agentic workflow engine

• Document upload, embedding, and search

• JWT authentication

• Free-tier hosting (Render, Vercel)

• S3-compatible file storage

• Automated tests & CI/CD

• Responsive UI using Tailwind CSS

• Admin Dashboard for Upload Documents, Documents List, Case Management, and User Management

1. Clone the repo

   git clone https://github.com/yourusername/semantic-case-manager.git
   cd semantic-case-manager

2. Run locally (dev)

   docker compose up

3. Set up TiDB Serverless

   • Create a free TiDB Cloud account: https://tidbcloud.com
   • Create a Serverless cluster
   • Note your connection string and set it in .env

4. Set up S3-compatible storage

   • Use Cloudflare R2 or any S3-compatible free-tier
   • Set credentials in .env

5. Configure environment variables

   • Copy .env.example to .env and fill in required values

/backend # FastAPI backend /frontend # React + Vite frontend

---


## Search Page

**Ask a Question:**
Type your question in the search bar. The app will search all cases or your selected case and provide a clear, organized answer based on the documents.

1. User Authentication

   • User visits the app and logs in via username/password (JWT)
   • Authenticated users are redirected to the dashboard.

2. Document Upload & Extraction

   • User uploads a document (PDF, DOCX, etc.).
   • Backend extracts text and stores file in S3-compatible storage.
   • Embeddings are generated and stored in TiDB.

3. Semantic Search

   • User enters a search query.
   • Backend performs vector search and returns semantically relevant response based on the documents/cases.

4. Case Management

   • User creates new cases, assigns/unassigns documents, and views case details.

5. Admin Dashboard

   • Admins access advanced features: user management, activity logs, and workflow triggers.

6. Workflow Automation

   • LLM-based summarization and notification workflows can be triggered on cases/documents.

MIT

   graph TD
   A[User] -->|Upload/Search| B[Frontend]
   B -->|API| C[FastAPI Backend]
   C -->|Vector Search| D[TiDB Serverless]
   C -->|File Storage| E[S3-Compatible Storage]
   C -->|Multi-Agent Workflow| H[Agentic Workflow Engine]
   H -->|LLM Calls| F[LLM/Webhook]
   H -->|Task Results| C
   H -->|Case/Doc Actions| D

• User Experience:
  • Add onboarding tooltips and contextual help for new users.
  • Improve error messages and loading indicators.
• Security:
  • Enforce stricter password policies and rate limiting.
  • Add audit logging for sensitive actions.
• Scalability:
  • Add support for multi-tenant organizations.
  • Implement background job queue for heavy tasks (embedding, LLM calls).
• Testing:
  • Increase test coverage for both backend and frontend.
  • Add end-to-end (E2E) tests for critical flows.
• DevOps:
  • Add monitoring/alerting for backend errors and downtime.
  • Automate database migrations on deploy.
• Documentation:
  • Expand API docs and add more usage examples.
  • Add a troubleshooting FAQ for common deployment issues.

• TiDB Serverless free-tier: 5GB storage, 20K QPS, 1M vector queries/month
• Example dataset: <100 docs, <1GB total
• S3/R2: free-tier limits apply

Health check:

curl http://localhost:8000/api/health
# Response: {"status": "ok"}

Login (JWT):

curl -X POST http://localhost:8000/api/auth/token -d "username=testuser&password=testpass"
# Response: {"access_token": ...}

User info:

curl -H "Authorization: Bearer <access_token>" http://localhost:8000/api/auth/me
# Response: {"username": "testuser","testpass"}

Admin endpoint:

curl -H "Authorization: Bearer <access_token>" http://localhost:8000/api/admin/ping
# Response: {"msg": "pong", ...}

Troubleshooting:

• If you see Could not validate credentials, check your token format and ensure the backend is running.
• If you see form data errors, ensure python-multipart is installed (see requirements.txt).

The major functionalities have been implemented and the project has been deployed to Vercel (frontend) and Render (backend).

Some improvement areas remain (see 'Improvement Areas & Recommendations' above), but the core features are complete and the app is deployed.