🇧🇩 BD Agent

Multi-Tool AI Agent for Bangladesh

An LLM-powered agent that answers real questions about Bangladeshi hospitals, institutions, and restaurants — grounded in real datasets, not hallucinations.

Overview • Demo • Architecture • Setup • Skills Demonstrated

---

📌 Overview

BD Agent is a tool-using AI agent built with LangChain that answers questions about Bangladesh by querying real structured data instead of relying purely on an LLM's memory. It combines:

• 🗄️ Three real-world datasets (hospitals, educational institutions, restaurants) ingested from HuggingFace into local SQLite databases
• 🧠 An LLM-based agent that reasons about which tool best answers a given question
• 🌐 A live web-search tool for general knowledge questions outside the datasets

The result: a domain-specific assistant that gives accurate, data-backed answers — a practical demonstration of retrieval-augmented, tool-using AI systems, one of the most in-demand patterns in applied AI/ML engineering today.

💬 Example Queries

> How many hospitals are in Dhaka?
> List hospitals with ICU facilities in Chittagong.
> What universities are listed in the institutions database?
> Show popular restaurants serving Bengali cuisine in Sylhet.
> What is the role of DGHS in Bangladesh?

The agent automatically routes each question to the correct tool — SQL query against the right database, or a live web search — without the user needing to specify which.

---

🏗️ Architecture

                         ┌──────────────────────┐
                         │     User Question     │
                         └──────────┬───────────┘
                                    │
                                    ▼
                       ┌────────────────────────┐
                       │   agent.py (LangChain)  │
                       │  LLM decides best tool  │
                       └────────────┬───────────┘
                                    │
          ┌─────────────┬──────────┼──────────┬─────────────┐
          ▼              ▼          ▼          ▼             
   ┌─────────────┐┌─────────────┐┌─────────────┐┌──────────────┐
   │ Hospitals   ││ Institutions││ Restaurants ││  Web Search  │
   │  SQL Tool   ││   SQL Tool  ││   SQL Tool  ││  (SerpAPI)   │
   └──────┬──────┘└──────┬──────┘└──────┬──────┘└──────────────┘
          ▼              ▼              ▼
   dbs/hospitals.db dbs/institutions.db dbs/restaurants.db
          │              │              │
          └──────────────┴──────────────┘
                          ▼
                  ┌───────────────┐
                  │ Final Answer  │
                  └───────────────┘

Data pipeline: ingest.py pulls each dataset from HuggingFace → converts to a pandas DataFrame → normalizes column names → writes to a local SQLite .db file via to_sql().

---

🗂️ Project Structure

bd_multi_tool_ai_agent/
├── agent.py                  # Entry point — builds & runs the LangChain agent
├── ingest.py                 # Downloads datasets & builds SQLite databases
├── tools/
│   ├── db_tools.py           # SQL tools: hospitals, institutions, restaurants
│   └── web_search_tool.py    # SerpAPI web-search tool
├── requirements.txt          # Python dependencies
├── .env.example               # API key template
└── .gitignore                 # Excludes secrets, DBs, venv

📊 Datasets

Source (HuggingFace)	Local Table
Mahadih534/Institutional-Information-of-Bangladesh	institutions
Mahadih534/all-bangladeshi-hospitals	hospitals
Mahadih534/Bangladeshi-Restaurant-Data	restaurants

Adding a new dataset is a one-line change — add a save_dataset_to_db(dataset_name, table_name, db_path) call in ingest.py.

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/delowarhossaincse63/bd_multi_tool_ai_agent.git
cd bd_multi_tool_ai_agent

2. Create a virtual environment

python -m venv venv
venv\Scripts\activate        # Windows
source venv/bin/activate     # macOS / Linux

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables

copy .env.example .env        # Windows
cp .env.example .env          # macOS / Linux

Then add your keys to .env:

OPENAI_API_KEY=your_openai_key_here
SERPAPI_API_KEY=your_serpapi_key_here

5. Build the databases

python ingest.py

Creates dbs/hospitals.db, dbs/institutions.db, dbs/restaurants.db.

6. Run the agent

python agent.py                                   # interactive mode
python agent.py "How many hospitals are in Dhaka?" # direct query

---

🧠 Skills Demonstrated

This project showcases practical, job-relevant experience with:

• LLM Agent Design — building tool-using agents with LangChain that reason about task routing
• Data Engineering — ETL pipeline: dataset ingestion, cleaning, normalization, and loading into SQLite
• API Integration — OpenAI API and SerpAPI, with secure key management via .env
• Database Design — schema normalization, SQL query tools, de-duplication logic
• Software Engineering Practices — modular code structure, .gitignore hygiene, secret management, reproducible setup
• Applied AI for Real-World Data — grounding LLM output in verified datasets to reduce hallucination

---

🔍 Verifying the Databases

sqlite3 dbs/hospitals.db "PRAGMA table_info('hospitals');"
sqlite3 dbs/hospitals.db "SELECT COUNT(*) FROM hospitals;"

Or in Python:

import sqlite3
conn = sqlite3.connect('dbs/hospitals.db')
cur = conn.cursor()
print(cur.execute("SELECT name FROM sqlite_master WHERE type='table';").fetchall())
print(cur.execute("SELECT count(*) FROM hospitals;").fetchone())
conn.close()

---

🛡️ Security Notes

The following are excluded from version control via .gitignore:

• .env — API keys
• dbs/*.db — generated database files
• venv/ / .venv/ — virtual environment
• __pycache__/, *.pyc

If a secret is ever committed accidentally:

git rm --cached .env
git commit -m "Remove .env from tracking"
git commit --amend --no-edit
git push -f origin main

---

🧯 Troubleshooting

Issue	Solution
OPENAI_API_KEY is required	Confirm .env exists and contains the key; restart the terminal.
GitHub blocks push (secret detected)	Remove the secret, amend the commit, force-push.
Agent gives unexpected answers	Check console logs for tracebacks; verify all keys in .env.
Web search not working	Confirm SERPAPI_API_KEY is set — without it, only SQL tools work.

---

🗺️ Roadmap

• Add more Bangladesh-specific datasets (transport, weather, government services)
• Build a lightweight web UI (Streamlit/FastAPI)
• Add caching for repeated queries
• Deploy as a public demo

---

🤝 Contributing

Contributions, issues, and feature requests are welcome. Feel free to check the issues page.

📄 License

This project is licensed under the MIT License.

📬 Contact

Delowar Hossain 📧 delowarhossain.cse.63@gmail.com 🔗 GitHub

---

⭐ If you find this project useful, consider giving it a star!