04 // projects

What I've Built

AI-powered, data-driven applications built from scratch with real metrics.

// 01 Jan 2026

Land Record NER System

Advanced Named Entity Recognition · NLP / Deep Learning

  • Production-grade BiLSTM-CRF NER system using PyTorch, enabling precise entity extraction from unstructured land records with zero rule-based dependencies.
  • Built a Combinatorial Grammar pipeline to synthesise 15,000+ training samples, enabling generalisation across complex linguistic patterns without manual annotation.
  • Implemented GloVe embeddings + targeted negative sampling to resolve edge cases like passive voice ambiguity and date/cost confusion in legal documents.
Python PyTorch BiLSTM-CRF GloVe Streamlit
⎇View on GitHub
// 02 Sept 2025

Electricity Bill Predictor

ML Web App · Appliance-based Consumption Forecasting

  • ML web app that predicts monthly electricity bills from appliance usage data — no per-device monitoring required, just household input patterns.
  • Collected & pre-processed a real-world dataset of Indian household appliances — handled missing values, normalisation, and feature engineering end-to-end.
  • Built a Streamlit dashboard with interactive visualisations using pandas, scikit-learn, and matplotlib for prediction analysis and energy insights.
Python scikit-learn Streamlit pandas matplotlib
⎇View on GitHub
// 03 Aug 2025

Cloud Cost Optimizer

AI Chatbot · Multi-cloud FinOps Platform

  • AI-driven cloud FinOps platform using Streamlit + Google Gemini for real-time conversational analysis and actionable cost-reduction strategies across AWS, Azure, and GCP.
  • Engineered a context-aware pipeline integrating pandas with LLMs — intelligently interprets uploaded cost reports and generates data-driven insights while optimising token usage.
  • Responsive interface with dynamic response stylising, real-time streaming, and session persistence for identifying idle resources across multi-cloud environments.
Streamlit Gemini API Python pandas LLMs
⎇View on GitHub