TaskPilot - Multi-Tool AI Agent

TaskPilot is a multi-tool AI agent that dynamically chooses between document retrieval, document summarization, web search, and direct reasoning using LangGraph.

TaskPilot is an AI agent that can answer questions using three complementary tools: RAG document retrieval, document summarization, and live web search. A LangGraph orchestrator dynamically decides which tool(s) to invoke for each query.

---

Overview

TaskPilot extends the Inflow AI RAG foundation into an agent workflow. Gemini decides whether a request needs one of the three tools, multiple tools, or no tool at all. FastAPI exposes the agent and document APIs, while React provides the Knowledge Base and chat experience.

Features

What is implemented:

• Clean production-style repository structure

• FastAPI backend with health endpoint (GET /api/health)

• Document upload and parsing (PDF, DOCX, TXT, MD)

• Recursive text chunking

• FAISS vector store and embedding pipeline

• Semantic similarity document search (POST /api/documents/search)

• SQLite metadata persistence layer

• Centralized Pydantic Settings configuration

• Structured logging system

• React + TypeScript + Vite frontend shell

• LangGraph StateGraph Agent (LLM-driven tool selection)

• RAG Retriever Tool (semantic search on selected documents)

• Document Summarizer Tool (map-reduce summarization)

• Web Search Tool (Tavily integration)

• Conversation Checkpointing (multi-turn memory via SQLite)

• Source Tracking & Tool Traces (isolated from internal reasoning)

• AgentService HTTP Bridge (POST /api/chat with persistence)

• Backend test suite covering health, RAG, agent, and chat APIs

• React Interface (responsive dark theme)

• Responsive Layout (Sidebar, document uploads, chat area)

• Markdown & Source Rendering (Rich chat bubbles with citations)

• Tool Traces (observable activity from backend responses)

• API integration hooks (useChat, useDocuments)

The current HTTP API is request/response based; the UI shows a truthful loading state until the response arrives.

---

Technology Stack

Layer	Technology
Frontend	React 19 · TypeScript 6 · Vite 8
Backend	Python 3.12+ · FastAPI · Uvicorn
AI Agent	LangGraph
AI Tools	LangChain · langchain-google-genai
RAG	FAISS · Sentence Transformers
Web Search	Tavily
Config	Pydantic Settings
Testing	pytest · pytest-asyncio

---

Project Structure

TaskPilot/
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI app factory + lifespan
│   │   ├── api/
│   │   │   ├── health.py        # GET /api/health
│   │   │   ├── documents.py     # Placeholder — Phase 2
│   │   │   └── chat.py          # Placeholder — Phase 3
│   │   ├── agent/               # LangGraph agent — Phase 3
│   │   ├── tools/               # RAG, Summarizer, WebSearch — Phase 2/3
│   │   ├── rag/                 # Vector store & retrieval — Phase 2
│   │   ├── services/            # Business logic layer
│   │   ├── models/
│   │   │   └── schemas.py       # Pydantic response models
│   │   ├── core/
│   │   │   ├── config.py        # Centralised Settings (pydantic-settings)
│   │   │   └── logging.py       # Logging configuration
│   │   └── utils/
│   ├── tests/
│   │   └── test_health.py       # Health endpoint tests
│   ├── requirements.txt
│   └── .env.example
│
├── frontend/
│   ├── src/
│   │   ├── components/
│   │   │   └── StatusBadge.tsx  # Animated status indicator
│   │   ├── services/
│   │   │   └── api.ts           # Centralised API client
│   │   ├── types/
│   │   │   └── api.ts           # TypeScript interfaces
│   │   ├── App.tsx
│   │   └── main.tsx
│   ├── vite.config.ts
│   └── .env.example
│
├── data/
│   ├── uploads/                 # User-uploaded documents (git-ignored)
│   └── vector_store/            # FAISS index (git-ignored)
│
├── .gitignore
├── README.md
└── ARCHITECTURE.md

---

Backend Setup

Prerequisites

• Python 3.12+
• pip

Steps

# 1. Create virtual environment
cd backend
py -3.12 -m venv .venv_new

# 2. Activate virtual environment
# Windows:
.venv_new\Scripts\activate
# macOS / Linux:
source .venv/bin/activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Copy and configure environment
cp .env.example .env
# Edit .env; health, docs, and document management work without provider keys

# 5. Run the development server
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Backend endpoints:

• Health check: GET http://localhost:8000/api/health
• Swagger docs: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc
• Documents: GET /api/documents, POST /api/documents/upload, DELETE /api/documents/{document_id}
• Document search: POST /api/documents/search
• Chat: POST /api/chat

Running Tests

cd backend
.venv\Scripts\python -m pytest tests/ -v   # Windows
# or:
python -m pytest tests/ -v                  # Unix with venv activated

---

Frontend Setup

Prerequisites

• Node.js 18+
• npm

Steps

# 1. Navigate to frontend directory
cd frontend

# 2. Install dependencies
npm install

# 3. Copy environment file
cp .env.example .env
# VITE_API_BASE_URL=http://localhost:8000 (default — change for production)

# 4. Start development server
npm run dev

Frontend runs at: http://localhost:5173

The Vite dev server proxies /api/* to the FastAPI backend automatically, so CORS is not an issue during development.

---

Environment Variables

Backend (backend/.env)

Variable	Default	Required	Description
APP_ENV	development	No	development / staging / production
GEMINI_API_KEY	(empty)	Optional for chat	Google Gemini API key
GEMINI_MODEL	gemini-3.6-flash	No	Gemini model identifier
TAVILY_API_KEY	(empty)	Optional for web search	Tavily API key
VECTOR_STORE_PATH	../data/vector_store	No	FAISS index storage path
UPLOAD_DIR	../data/uploads	No	Document upload directory
FRONTEND_ORIGIN	http://localhost:5173	No	CORS allowed origin
MAX_UPLOAD_SIZE_MB	20	No	Max upload file size
RAG_CHUNK_SIZE	1000	No	Token chunk size for text splitting
RAG_CHUNK_OVERLAP	200	No	Chunk overlap
RAG_TOP_K	5	No	Top-k documents to retrieve

Frontend (frontend/.env)

Variable	Default	Description
VITE_API_BASE_URL	http://localhost:8000	FastAPI backend base URL

---

Example prompts

• "What does the uploaded project report say about the evaluation?"
• "Summarize this document in bullet points."
• "What is the latest public information about ...?"

Without GEMINI_API_KEY, document management and health/docs remain available and chat returns a clear configuration error.

More guided scenarios are in DEMO.md. Interview preparation is in INTERVIEW_GUIDE.md.

Architecture

flowchart LR
	UI[React UI] --> API[FastAPI]
	API --> SERVICE[AgentService]
	SERVICE --> GRAPH[LangGraph StateGraph]
	GRAPH -->|LLM requests tool| TOOLS[Tool node]
	TOOLS --> RAG[Document Retriever]
	TOOLS --> SUM[Document Summarizer]
	TOOLS --> WEB[Web Search]
	TOOLS --> GRAPH
	GRAPH --> RESPONSE[ChatResponse]
	RESPONSE --> UI

Loading

Limitations and future improvements

• SQLite and local FAISS are intended for local or small deployments.
• Gemini is required for agent chat; Tavily is required for live web search.
• There is no authentication or multi-user workspace isolation.
• Chat is request/response based rather than streamed.
• Natural next steps are streaming, authentication, cloud vector storage, and multi-user workspaces.

Development

Both servers must be running simultaneously for the full stack:

# Terminal 1 — Backend
cd backend
.venv_new\Scripts\activate   # or: source .venv_new/bin/activate
uvicorn app.main:app --reload --port 8000

# Terminal 2 — Frontend
cd frontend
npm run dev