LangChain RAG Chatbot (LLM-Powered Prompt System)

This is a demonstration of my applied skills in using LLMs for structured data extraction, dynamic prompt creation, and full-stack delivery. It integrates prompt engineering, schema-aware NER, and real-time LLM interaction via a Flask web app.

---

✅ Highlights (AI Engineer / Data Scientist / Researcher Perspective)

• 🔍 Schema-Aware NER with LLMs: Custom prompt templates exatract relevant fields like domain, method, framework, etc. using GPT models. Easily extendable for research/data pipelines.
• 🧠 LLM-Driven Prompt Generation: Extracted fields populate prompt templates programmatically—core to building adaptive, context-rich LLM systems.
• 🤖 OpenAI Integration: Works with gpt-3.5-turbo, gpt-4o, modular for any provider (Anthropic, Mistral, Ollama, etc.).
• 🌐 Flask-based Frontend: Deployed as a local web interface—serves as a real-world UX for AI prototypes.
• 📦 Pinecone Integration: Vector DB is setup-ready for Retrieval Augmented Generation (RAG)—can scale with document ingestion.
• 📊 LLM Output Logging: Logs full trace of extracted fields, generated prompt, and output for downstream eval/debug.

---

🧠 Tech Stack

• Python, Flask: Web backend
• LangChain: Prompt orchestration and chains
• LangChain OpenAI / Community: Model connectors
• Pinecone: Vector DB (RAG-ready)
• Dotenv: Environment handling
• HTML / Jinja: Simple UI rendering

---

🚀 Run Locally

git clone <your-repo-url>
cd <project-directory>
pip install -r requirements.txt

Create a .env file:

OPENAI_API_KEY=your_openai_key
PINECONE_KEY=your_pinecone_key
LANGCHAIN_TRACING_V2=true

Run the app:

python app.py

Open http://localhost:5000

---

🔬 Example Use Case (LLM Evaluation)

Input:

"How can I use RAG with LangChain for document QA?"

Extracted Fields:

{
  "domain": "AI",
  "method": "RAG",
  "framework": "LangChain"
}

Prompt Sent to LLM:

SYSTEM: You are a helpful assistant. Use available context to answer precisely.
USER: How can I use RAG with LangChain for document QA?

Bot Reply:

"To use RAG in LangChain..."

---

📌 Key Code Components

🧩 DataIngestionLoader

• Handles both file and URL input
• Supports .pdf, .txt, .csv, .json, .docx, .html, and web pages
• Internally uses LangChain’s document loaders

🧠 SchemaNER

• LLM-powered schema-based entity extraction using OpenAI
• Returns structured JSON fields used in prompts

🧾 PromptBuilder

• Dynamically injects extracted fields into a system/user prompt
• Enables context-rich, personalized responses

---

💼 Why This Matters

This project demonstrates:

• 🛠 How to build LLM applications with structured input/output
• 🧠 How to extract interpretable, reusable metadata using schema prompts
• ⚙️ How to combine prompt engineering, context injection, and LangChain pipelines
• 📊 How to log and evaluate generation quality using prompt traces
• 🚀 How to set up full-stack AI workflows (LLM + web UI + document ingestion)

---

🧱 Planned Additions

• Document chunking + vector embeddings (full RAG)
• LangSmith or Phoenix tracing integration
• Streamlit/Gradio UI version
• PDF/JSON export of results
• API endpoints for backend-only integration

---

📬 Contact

Feel free to reach out if you're looking for a hands-on AI Engineer / Researcher with:

• Experience building LLM applications from scratch
• Knowledge of prompt + context engineering
• Practical backend + frontend integration skills
• A passion for building real-world, user-facing AI tools

---

📄 License

MIT

---

Built with ❤️ using LangChain, Flask, OpenAI, and too many cups of ☕.