I'm an engineer focused on Applied AI, Local LLMs, and Agentic RAG systems. I enjoy building production-oriented applications that run efficiently on consumer hardware without compromising on privacy.
Currently building Wayfarer, an AI-driven job search platform, and LocalBrainNotes, a local-first agentic assistant. I specialize in end-to-end AI product development—from local inference and RAG pipelines to full-stack deployment with Docker and React.
- 🔭 Building: Full-stack AI platforms (LangGraph, ChromaDB, FastAPI, React/TypeScript).
- 🧠 Focus Areas: Agentic AI, Multi-provider LLM Routing, Local Inference (Ollama), MLOps.
- 🏆 Achievements: Rank 158/4,506 (Top 3.5%) in Kaggle Hull Tactical Competition.
- 🌱 Learning: CUDA acceleration, Multi-agent orchestration, evaluation frameworks, on-device model optimization, and advanced prompt engineering.
- 🎯 Looking For: AI Engineer / Founding Engineer roles (Bengaluru | Remote)
AI & Machine Learning:
Backend & Infrastructure:
Frontend:
A full-stack AI platform that parses resumes, ranks live job posts, and provides ATS-oriented feedback with human-in-the-loop review.
- Key Features: Multi-provider LLM routing (NVIDIA NIM → Ollama), Redis-backed caching, and a custom resume graph for token-efficient matching.
- Stack: FastAPI, React/TypeScript, ChromaDB, Docker, Ollama.
- View Project | Live Demo (Docs)
A privacy-first, Obsidian-compatible assistant that runs 100% locally. It uses a LangGraph agent loop for complex reasoning over markdown vaults.
- Key Features: Agentic RAG (Router → Retrieve → Grade → Generate), local Ollama inference, and a Tauri desktop shell.
- Stack: LangGraph, ChromaDB, Ollama, Python, Tauri.
- View Project
An offline video-automation platform that extracts high-value moments from long-form content using LLM analysis and face-tracking.
- Key Features: GPU-accelerated media pipelines (NVENC/CUDA), WhisperX for word-level transcription, and semantic clip deduplication.
- Stack: Ollama, Gemma 3, WhisperX, PyQt6, FFmpeg, CUDA.
- View Project
Quantitative asset-allocation solution for the Hull Tactical Market Prediction challenge using Powell optimization and Sharpe-ratio maximization.
- Key Features: Modified Sharpe-ratio maximization with volatility penalties, leverage bounds, and risk constraints. Rank 158/4,506 globally.
- Stack: Python, NumPy, Powell optimization, Jupyter Notebook.
- View Notebook
- Top 3.5% globally in Kaggle Hull Tactical Competition (Rank 158/4,506).
- Published Research on Customer Segmentation using K-Means (IJNRD).
- Microsoft Certified in Data Engineering on Azure.
