Maahir Garg· MG

§ Projects

Projects by Maahir Garg. Things I've made, broken, re-made.

A running archive. Some shipped, some sat in a drawer, some became the curriculum. Click in for the notes.

Filter:
  1. 01 / 09Nov 2025

    Optimizing BERT for Question Answering

    Achieved a 69% reduction in model size (440MB to 128MB) with <0.5% F1 loss via post-training quantization. Implemented custom mask-enforced pruning to reach 54.7% sparsity and analyzed efficiency vs generalization trade-offs.

    BERTQuantizationPruningNLP
  2. 02 / 09Oct 2025

    Domain-Specific LLM Reasoning with PEFT

    Built a PEFT pipeline (BitFit + few-shot) for health insurance claim reasoning, fine-tuning <0.1% of parameters (bias terms only). Improved reasoning accuracy from 47% to 73% on clause-extracted scenarios.

    LLMPEFTBitFitFew-Shot Learning
  3. 03 / 09Dec 2025

    Multimodal Hand Tracking for Stroke Rehab

    Patent-pending framework that unifies iPhone and Apple Vision Pro into a single tracking surface for hand-eye coordination assessment in stroke rehabilitation. Co-invented at Interactive 3D Lab; built in Swift and RealityKit; validated in controlled studies with healthcare practitioners at 89% cross-device accuracy. Research paper in preparation.

    SwiftRealityKitVision ProiPhone
  4. 04 / 09Ongoing

    LeetCoding

    An ongoing LeetCode practice track focused on deepening algorithmic thinking and problem‑solving speed, backed by live stats.

    AlgorithmsData StructuresProblem SolvingPython
  5. 05 / 09Sep 2024

    StackExchange Scraping

    Data pipelines to analyze Stack Overflow feedback mechanisms.

    PythonData EngineeringWeb Scraping
  6. 06 / 09Mar 2024

    Google Scholar Scraping

    Automated scraping tool for academic research analysis, saving 520+ hours.

    PythonPostmanAutomation
  7. 07 / 09Mar 2024

    CAAQMS Web Scraping

    Air quality data scraper using SQLite and JWT, saving 10,000+ hours.

    PythonSQLiteJWT
  8. 08 / 09Mar 2024

    Llama2 Fine-tuning

    End-to-end pipeline for fine-tuning Llama2 large language models using LoRA / PEFT adapters. Covers data preparation, supervised fine-tuning, and evaluation - built as a reproducible notebook for experimenting with domain adaptation of open-weight LLMs.

    PythonLlama2LoRAPEFT
  9. 09 / 09Apr 2026

    Portfolio Website

    This site. Editorial field-notebook redesign: warm paper canvas and oxidized-red pencil mark, indexed sections, live Singapore clock, both light and dark tuned independently. Built with Next.js 16 + Tailwind v4; motion respects prefers-reduced-motion; a11y basics from skip-link to keyboard lightbox.

    Next.jsReactTypeScriptTailwind CSS v4