I build things at the intersection of machine learning systems, computer vision, and production-grade AI. Currently pursuing my Master's in Computer Science at Seattle University, with a background in AI and Data Science from Vishwakarma University (Pune, India).
I care about understanding systems deeply — not just using them. That means implementing attention mechanisms from scratch, building RAG pipelines without LangChain, and engineering real security layers into ML applications.
🎬 WatchWise · AWS Hacks 2026
AI-powered streaming recommendation engine built during AWS Hacks 2026.
🔍 VibeSearch · repo
Multi-modal fashion discovery engine — upload an image, describe a style vibe, get ranked visual matches. OpenCLIP + FAISS + Gemini 2.5 Flash, with a 3-layer security module and MLflow experiment tracking.
🧠 LLM Systems Engineering · repo
Production-grade QLoRA fine-tuning pipeline + adversarial RAG system with prompt injection defense. Built from scratch — no LangChain, no shortcuts.
| Project | What it is | Stack |
|---|---|---|
| Mixtral MoE from Scratch | Sparse MoE Transformer replicating Mixtral 8x7B's core architecture | PyTorch |
| Grouped-Query Attention | GQA from scratch — the attention mechanism in Llama 3 & Mistral; 4x KV-cache reduction | PyTorch |
| Re-Pair Text Compressor | BPE tokenizer foundation; scales to 1M characters with O(1) deletions | Python |
| Plant Leaf Classification | 99.65% accuracy leaf ID with VGG16 transfer learning | TensorFlow |
| Lexical Analyser | Full NLP pipeline with GUI — tokenization, POS tagging via Stanford CoreNLP | Python |
ML / AI
PyTorch TensorFlow HuggingFace PEFT OpenCLIP FAISS Sentence-Transformers VGG16 ResNet50
LLM & RAG
QLoRA RAG HyDE Prompt Injection Defense Fine-Tuning MLflow
Backend & Data
Python Flask FastAPI Streamlit MySQL REST APIs Power BI
Tools
Git Google Colab Jupyter Figma Google Workspace
- Mint Leaves Dataset for Condition Analysis and ML Applications
- Intelligent Predictive Maintenance for Smart Building Systems
- Applications of ML and AI in Metal Die Casting — ALUCAST Journal
Seattle, WA · Open to internships & research collaborations