🤖 Portfolio Chatbot — RAG-based Personal AI Assistant

🔗 Live Demo: https://chatbot.zid.web.id/

---

📌 Overview

A context-aware AI chatbot designed to transform a traditional portfolio into an interactive conversational interface 💬

Instead of reading static CVs, users can directly ask about:

• Experience
• Skills
• Projects

The system responds with relevant, data-driven answers using a lightweight RAG (Retrieval-Augmented Generation) approach.

---

⚙️ Architecture

User Input
   ↓
Intent Detection
   ↓
Topic Mapping
   ↓
Context Retrieval (data.txt)
   ↓
Prompt Injection
   ↓
OpenAI API
   ↓
Response (UI)

---

🧠 Core Features

• 💬 Clean & Responsive UI
• 🧠 Lightweight RAG system
• 🎯 Keyword-based topic routing
• 📂 Structured knowledge base (section-based)
• 🔍 Portfolio relevance scoring
• 🧵 Conversation history support
• ⚡ FastAPI backend (fast & efficient)

---

🔬 Technical Breakdown

📥 Context Retrieval

Only relevant data is injected into the model:

context = retrieve_context(user_input)

✔ Improves accuracy ✔ Reduces token usage ✔ Minimizes hallucination

---

🏷️ Topic Routing

KEYWORD_TOPIC → TOPIC_MAP → SECTIONS

Maps user queries into structured knowledge domains.

---

📂 Knowledge Structure

Stored in:

data.txt

Format:

==== SECTION NAME ====
content...

---

🔍 Portfolio Ranking

score += 1 if word in text else 0

Ensures only relevant projects are returned.

---

🧩 Prompt Injection

system_prompt = BASE_PROMPT + context

Constrains the model to grounded responses.

---

🖥️ Tech Stack

Backend

• FastAPI
• Python
• OpenAI API

Frontend

• TailwindCSS
• Vanilla JavaScript
• Lucide Icons

---

📁 Project Structure

.
├── index.html        # UI
├── main.py           # Backend
├── data.txt          # Knowledge base
├── prompt.txt        # Prompt config
├── .env              # API key

---

🔌 API Endpoint

POST /chat

{
  "message": "What are your skills?",
  "history": []
}

---

🎯 Design Decisions

• File-based knowledge → simple & portable
• Keyword routing → deterministic & lightweight
• No embeddings → avoid overengineering
• Limited history → token efficiency

---

🚀 Future Improvements

• Embedding-based retrieval
• Vector database (FAISS / Pinecone)
• Streaming responses
• Multi-language support
• Docker deployment

---

👤 Author

Fida Yuzida

---

⭐ Final Thought

Turning a portfolio into a conversation creates a more engaging way to explore a candidate 🚀