DBEB Agent

Unified AI Agent for document intelligence built with FastAPI + LangGraph backend and Next.js 15 frontend.

Key Features

• Unified AI Agent: Single chat interface that automatically routes to the right workflow based on your intent.
• Smart Intent Classification: The agent understands whether you want to chat, ingest documents, or evaluate candidates.
• RAG-Powered Chat: Retrieval-augmented responses using LangGraph orchestration and Gemini Flash.
• Document Ingestion: Add PDFs to the permanent knowledge base through natural conversation.
• Candidate Evaluation: Upload CSV + resumes ZIP and get automated screening with LLM-backed reasoning.
• Streaming Responses: Real-time token streaming for all operations.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        Frontend (Next.js)                        │
│                     Single Page AI Agent UI                      │
└─────────────────────────────────────────────────────────────────┘
                                 │
                                 ▼
┌─────────────────────────────────────────────────────────────────┐
│                    POST /agent (Unified Endpoint)                │
├─────────────────────────────────────────────────────────────────┤
│  1. Receive message + files                                      │
│  2. Classify intent (heuristics + LLM fallback)                  │
│  3. Route to handler:                                            │
│     ├─ chat → RAG pipeline with LangGraph                        │
│     ├─ ingest → Add documents to Qdrant                          │
│     └─ evaluate → Batch candidate screening                      │
│  4. Stream results back                                          │
└─────────────────────────────────────────────────────────────────┘

Project Layout

backend/app/
├── main.py                  # FastAPI entrypoint
├── api/
│   ├── agent_endpoint.py    # Unified /agent endpoint
│   └── endpoints.py         # Legacy endpoints
└── services/
    ├── agent_router.py      # Intent classification
    ├── llm.py               # LangGraph workflow
    ├── evaluator.py         # Candidate evaluation
    └── vector_store.py      # Qdrant integration

dbebui/                      # Next.js single-page agent UI

Requirements

• Python 3.10+
• Node.js 18+
• Qdrant instance (local or remote)
• Google Generative AI API key (GOOGLE_API_KEY)

Backend Setup

cd backend
python -m venv .venv
source .venv/bin/activate  # or .venv\Scripts\activate on Windows
pip install -r ../requirements.txt
uvicorn backend.app.main:app --reload --port 8000

Environment variables in backend/.env:

GOOGLE_API_KEY=...
ADMIN_KEY=...
QDRANT_URL=http://localhost:6333
QDRANT_API_KEY=

API Endpoints

Primary Endpoint

POST /agent — Unified AI agent endpoint

Form fields:

• message: Your text prompt
• thread_id (optional): Session ID for conversation continuity
• files (optional): One or more file uploads

The agent classifies your intent and routes to:

• chat: RAG-powered Q&A
• ingest: Add documents to knowledge base
• evaluate: Batch candidate screening

Response: Server-Sent Events stream with intent, token, results, done events.

Legacy Endpoints

• POST /stream — Direct chat stream
• POST /upload — Admin PDF ingestion (requires X-Admin-Key header)
• POST /evaluate-candidates — Direct candidate evaluation

Frontend Setup

cd dbebui
npm install
npm run dev

Open http://localhost:3000 to use the unified agent interface.

Usage

Chat / Q&A

Just type your question. The agent will search the knowledge base and respond.

Ingest Documents

Upload a PDF and say "Add this to the knowledge base" or "Ingest this document".

Evaluate Candidates

Upload three files:

1. Criteria document (PDF/TXT): Selection requirements
2. Candidates CSV: Must include resume_filename column
3. Resumes ZIP: Archive containing the resume files

Then say "Evaluate these candidates" and the agent will process each one.