I bridge the gap between research-grade ML models and production hardware. My focus is on air-gapped multimodal agents, medical diagnostic pipelines, and high-concurrency streaming systems.

Local NotebookLM & Mind-Mapper

Built a local research assistant using LLM-aided OCR to extract handwritten "stickers" and notes. Powered by a local LLM, it dynamically generates interactive, expandable flowcharts and a vectorized knowledge base for local RAG. ⚙️ Flow: Image Ingest ➔ LLM OCR ➔ Entity Extraction ➔ Dynamic Expanding Flowchart 🛠️ Tech: Local LLM Dynamic UI OCR Local RAG

Nebulai Ecosystem

Architected a local LLM cluster and an agentic routing system. Cut token costs by 30% and operational costs by 60%. ⚙️ Flow: User Query ➔ Router Agent ➔ vLLM Cluster ➔ Cost-Optimized Output 🛠️ Tech: vLLM Agentic Orchestration Cost Optimization

OpTomo (Medical Diagnostic Pipeline)

Designed an end-to-end inference pipeline for breast cancer detection, migrating heavy Python logic to hardware-accelerated bindings for a 40% latency reduction. ⚙️ Flow: Medical Scan ➔ Python Preprocessing ➔ C++ Inference Bindings ➔ Real-Time Diagnosis 🛠️ Tech: C++ Python Hardware Optimization Edge Inference

Secure Enterprise RAG

Built an air-gapped "Talk-to-your-Data" tool utilizing hybrid search and robust data engineering pipelines. ⚙️ Flow: Enterprise Data ➔ Apache NiFi Ingest ➔ Hybrid Search (BM25 + Vector) ➔ Secure LLM 🛠️ Tech: Apache NiFi Milvus / Pinecone Hybrid Search Air-gapped LLM

Mobile Edge-AI Engine

Engineered a hot-swappable mobile inference engine capable of switching neural architectures at runtime without requiring app store updates. ⚙️ Flow: Android App ➔ Kotlin Orchestrator ➔ ONNX Model Hot-Swap ➔ On-Device Inference 🛠️ Tech: Kotlin ONNX Android SDK Neural Architecture

Edge AI & Hardware Optimization

Generative AI, Agents & Audio

Data Engineering, RAG & Databases

Computer Vision, Analysis & Visualization

Full Stack & DevOps