Portfolio Bot 🤖

A RAG-powered portfolio assistant that ingests your personal data (PDF, DOCX, TXT, Markdown) and answers questions about you through a sleek streaming chat UI.

Architecture

my_data/  ──▶  Chunking  ──▶  NVIDIA Embeddings  ──▶  pgvector (PostgreSQL)
                                                            │
User Question  ──▶  Embed Query  ──▶  Cosine Search  ──────┘
                                           │
                                     Scope Guard  ──▶  (reject off-topic)
                                           │
                                     Groq LLM (streaming)  ──▶  Chat UI

Quick Start

1. Prerequisites

• Python 3.11+
• PostgreSQL with the pgvector extension installed
• NVIDIA NIM API key (for embeddings) — build.nvidia.com
• Groq API key (for LLM) — console.groq.com

2. Set up PostgreSQL

-- In psql:
CREATE EXTENSION IF NOT EXISTS vector;

3. Configure environment

Edit .env in the project root with your credentials:

NVIDIA_API_KEY=nvapi-...
GROQ_API_KEY=gsk_...
PGPASSWORD=your_pg_password

4. Install dependencies

pip install -r requirements.txt

5. Add your data

Drop your files into app/my_data/:

app/my_data/
├── resume.pdf
├── about_me.md
├── projects.txt
└── experience.docx

Supported formats: .pdf, .docx, .txt, .md

6. Run

uvicorn app.main:app --reload

Open http://localhost:8000 in your browser.

How It Works

1. Ingestion — On startup (and when files change), documents are chunked and embedded via NVIDIA NIM, then stored in pgvector.
2. Query — User questions are embedded and cosine-searched against stored chunks.
3. Scope Guard — If the closest chunk is too distant (configurable threshold), the bot rejects the question as out-of-scope. The system prompt further constrains answers.
4. Streaming — Groq streams the answer token-by-token via SSE to the browser.

API

POST /api/chat

{
  "question": "What are their skills?",
  "top_k": 5
}

Returns text/event-stream:

data: {"token": "They "}
data: {"token": "have "}
data: {"token": "experience..."}
data: [DONE]

Project Structure

portfolio-bot/
├── app/
│   ├── main.py              # FastAPI app + lifespan
│   ├── config.py             # Pydantic Settings
│   ├── watcher.py            # Watchdog auto-ingest
│   ├── my_data/              # YOUR data goes here
│   ├── api/routes/chat.py    # SSE streaming endpoint
│   ├── services/
│   │   ├── retrieval.py      # RAG pipeline + scope guard
│   │   └── ingestion/
│   │       ├── chunking.py   # File readers + text splitting
│   │       └── embedding.py  # NVIDIA embeddings + pgvector storage
│   └── static/index.html     # Chat UI
├── .env                      # Configuration
├── requirements.txt
└── README.md