"There's something magical about systems that can think, remember, and adapt."
My evolution as a builder spans from foundational algorithmic implementations to production-scale AI systems. Every project I build is designed to bridge complex frameworks, optimize performance, and solve real-world system bottlenecks through local-first and zero-trust principles.
| Category | Technologies |
|---|---|
| Languages |       |
| AI / ML & Vector Engines |       |
| Backend & Infrastructure |       |
My latest work combining AI, infrastructure, and creative engineering.
| Project | Focus Area | Description |
|---|---|---|
| Agent-Protocol-Bridge | Framework Interop | A framework that unifies MCP tools, A2A agents, LangChain, and LangGraph into one coherent system. Solves the framework fragmentation problem so you don't have to choose sides. |
| retainr | Python, FAISS, SQLite | A persistent, queryable memory library for AI apps—local-first with zero API cost. Provides semantic recall, memory decay, importance scoring, and GDPR-friendly deletion in one line of code. |
| Aegis Proxy | Privacy & Security | A privacy-first gateway between your app and LLM APIs. Detects and anonymizes PII before it leaves your infrastructure using Go, FastAPI, and Redis for zero-trust access. |
The stuff I'm actively polishing and shipping.
| Project | Tech Stack | Description |
|---|---|---|
| RARE: Agentic Retrieval | Reasoning AI | Experiments in making reasoning models smarter by pairing them with agentic retrieval. DeepSeek + smart context = higher accuracy answers. |
| GENAI_Prem-prom | LangGraph, Twilio, LLMs | A bilingual (English/Hinglish) multi-agent voice survey system for hospital patient feedback. Features intelligent empathy feedback generation and department-specific question routing. |
| Bilateral-Visioner (BICAL-Net) | PyTorch, MobileViT, YOLO | Medical AI for pulmonary disease detection using dual-stream bilateral cross-attention. SimCLR pre-training → YOLO segmentation → Grad-CAM explainability with 99% accuracy. |
Projects that taught me algorithms, system design, and how to actually build things.
| Project | Domain | Description |
|---|---|---|
| GUI-Driven-Text-Compression | Algorithms | Implemented Huffman coding, LZ77, and LZW from scratch with a GUI. Real data structures, real algorithms, real understanding in Java/Python. |
| Maze-Solver via Q-Learning | Reinforcement Learning | Built a physical robot that learns to navigate mazes using reinforcement learning. ML theory meets hardware, with a laptop sending instructions in real-time. |
| LLM for Hydrogen Storage | Domain-Specific AI | A domain-specific LLM for extracting insights from research papers. Performs summarization, QA, and knowledge synthesis for scientific literature. |
Feel free to connect with me on LinkedIn