NSD Alumni Intelligence Pipeline

A production-ready Python pipeline for automatically collecting and structuring publicly available information about National School of Drama (NSD) graduates (2014–2019).

The system takes a list of NSD alumni names and graduation years, searches the internet, extracts relevant information using a local LLM via Ollama, and generates a structured database.

---

Features

Input Support

• CSV input
• Excel (.xlsx) input

Expected columns:

Name	Graduation Year
Amanpreet Kaur	2014
Chirag Garg	2014

---

Information Collected

Basic Information

• Full name
• Alternative names
• Graduation year
• Current profession
• Current organization
• Current city/location

Online Presence

• Personal website
• Portfolio website
• LinkedIn
• Instagram
• Facebook
• YouTube
• IMDb
• Twitter/X
• Other public profiles

Theatre Information

• Plays acted in
• Plays directed
• Plays written
• Theatre groups
• Theatre companies
• Repertory associations
• Film appearances
• TV appearances
• Workshops conducted
• Awards

Public Contact Information

Only publicly available information is collected:

• Public phone numbers
• Public email addresses
• Public contact pages
• Management contacts

---

Ethical Constraints

This project intentionally avoids:

• Guessing phone numbers
• Guessing email addresses
• Inferring personal information
• Accessing private content
• Bypassing logins
• Circumventing website restrictions
• Collecting non-public data

---

Search Sources

The pipeline searches across multiple publicly available sources:

• Google/DuckDuckGo
• LinkedIn
• Instagram
• Facebook pages
• IMDb
• Theatre websites
• Festival websites
• News articles
• YouTube interviews
• Portfolio websites
• Personal websites
• NSD pages
• Repertory companies
• Medium articles
• Public PDFs
• Wikipedia

---

Project Structure

project_root/
│
├── input/
│   └── nsd_graduates.xlsx
│
├── output/
│   ├── alumni_data.csv
│   ├── alumni_data.xlsx
│   └── alumni_data.json
│
├── cache/
│
├── logs/
│
├── models/
│
├── main.py
├── search.py
├── extractor.py
├── llm_parser.py
├── utils.py
├── requirements.txt
│
└── README.md

---

System Architecture

Input CSV/Excel
        ↓
Generate contextual search queries
        ↓
Collect search results
        ↓
Extract webpage text
        ↓
Send data to local LLM
        ↓
Parse structured JSON
        ↓
Confidence filtering
        ↓
Deduplication
        ↓
Output generation

---

LLM Configuration

This project uses a local LLM through Ollama.

Default:

gemma3

Fallback:

llama3:8b

Install Ollama:

https://ollama.com/download

Pull model:

ollama pull gemma3

or

ollama pull llama3:8b

Test:

ollama run gemma3

---

Installation

Clone repository:

git clone https://github.com/yourusername/nsd-alumni-intelligence.git

cd nsd-alumni-intelligence

Create virtual environment:

Windows:

python -m venv venv

venv\Scripts\activate

Linux/Mac:

python3 -m venv venv

source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

---

Example requirements.txt

pandas
openpyxl
requests
beautifulsoup4
duckduckgo-search
ollama
tqdm
retry
loguru
rapidfuzz
lxml

---

Input Format

Place your Excel or CSV file inside:

input/

Example:

Name,Graduation Year
Amanpreet Kaur,2014
Chirag Garg,2014
Jayanta Narzary,2014

---

Running the Project

Run:

python main.py

The pipeline will:

1. Read names
2. Generate contextual search queries
3. Search the web
4. Download webpage content
5. Extract structured information
6. Run local LLM analysis
7. Generate final outputs

---

Output

Generated files:

output/
├── alumni_data.csv
├── alumni_data.xlsx
└── alumni_data.json

Example output:

{
  "name":"Amanpreet Kaur",
  "graduation_year":"2014",
  "current_profession":"Actor",
  "current_location":"Mumbai",
  "website":null,
  "portfolio":null,
  "linkedin":"...",
  "instagram":"...",
  "plays_acted_in":[
      "Play A",
      "Play B"
  ],
  "plays_directed":[
      "Play C"
  ],
  "public_emails":[
      "contact@example.com"
  ],
  "sources":[
      "https://..."
  ]
}

---

Robustness Features

Reliability

• Retry handling
• Request timeouts
• Network failure handling
• Logging

Data Quality

• Duplicate detection
• Ambiguous name handling
• Confidence scoring
• Low-confidence filtering
• Structured JSON validation

Performance

• Caching
• Rate limiting
• Reduced duplicate searches

---

Logs

Logs are stored in:

logs/

Example:

2026-05-26 18:20:41 INFO Searching: Amanpreet Kaur NSD
2026-05-26 18:20:44 INFO Parsing results
2026-05-26 18:20:48 INFO Confidence Score: 0.91

---

Running in VS Code

Open folder:

code .

Select interpreter:

Ctrl + Shift + P
Python: Select Interpreter

Choose:

venv

Run:

python main.py

---

Future Improvements

• Parallel processing
• GPU acceleration
• Vector database integration
• RAG-based retrieval
• Semantic duplicate detection
• Theatre-specific source prioritization
• Dashboard interface

---

Disclaimer

This tool only collects publicly available information.

Users are responsible for complying with website terms of service, privacy laws, and ethical data usage practices.

---

License

MIT License