Developed a privacy-first AI tool that generates interactive mind maps entirely offline using local open-source language models. The system accepts multiple input formats (PDFs, Word, text, and website content), extracts key concepts, and visualizes them dynamically. Optimized performance for CPU-only environments, integrating automated parsing, concept-limiting strategies, and clean layout generation. Overcame challenges of limited tokens and slower processing speed while demonstrating that powerful, intelligent AI applications can be built without reliance on cloud APIs. Applications span learning, brainstorming, research, and knowledge management.

Developed an AI-powered question auto-completer with a custom LSTM model trained on 77,533 developer questions, achieving accuracy in predicting the next word's context. Created a data preprocessing pipeline, converting raw Code Review Stack Exchange data into JSON, reducing data preparation time for model training.

Developed an AI-based IT Services Bot using Kore.ai, managing simulated interactions for password resets, issue reporting, and equipment ordering during training. Implemented an issue reporting workflow, capturing employee ID and issue descriptions, enabling structured issue tracking with 95% accuracy in user input recognition.

Successfully built and demonstrated an innovative solution to automate the SDLC using IBM Watsonx.ai services. Competed against 70+ teams and 200+ participants at LTIMindtree’s </HACK> THE FUTURE 2.0 event, conducted in collaboration with IBM, and won a special prize for innovation and excellence.

Collected 500+ domain-specific medical question-answer pairs, creating a high-quality dataset for BioGPT fine-tuning. Optimized Microsoft’s BioGPT on Kaggle’s free CPU resources, demonstrating a 30% decrease in compute needs and maintaining high model relevance.