M.Tech @ DA-IICT · Turning research papers into agentic systems.
- 🟢 Now: Contributing to agno — open-source agent framework
- 🔧 Just shipped: Odoo MCP Server with Langfuse observability
- 🧪 Exploring: Context personalization in OpenAI Agents SDK + remote MCP architectures
- 🎓 M.Tech in Machine Learning @ DA-IICT (Dhirubhai Ambani University formerly known as DA-IICT)
- 🤖 Building agentic systems — MCP servers, OpenAI Agents SDK, multi-agent orchestration, AGNO, LangChain, LangGraph, LangFuse
- 🔬 Implementing research papers — Transformer (Vaswani et al.), Micrograd, SISA Unlearning, PaliGemma
- ✍️ Writing on Medium, mentoring on Topmate
- 🤖 AI Agents & MCP — agent orchestration, tool-use, remote MCP servers
- 🔍 Retrieval-Augmented Generation — production-grade RAG techniques
- 🧠 Foundation Models — fine-tuning, vision-language, machine unlearning
- 🚀 End-to-end ML Systems — from research papers to deployed services
Production MCP server (built on Anthropic's Model Context Protocol) bridging Odoo ERP to LLM agents, instrumented with Langfuse for tracing. Stack: MCP · OpenAI Agents SDK · Langfuse · Odoo
AI safety / alignment work — Sharded, Isolated, Sliced, Aggregated training lets models forget specific data on demand. From the SISA paper, with a YouTube walkthrough. Stack: PyTorch · Python · Research-paper implementation
Re-implementing Google's PaliGemma multimodal architecture — image understanding + language generation, from scratch in PyTorch. Stack: PyTorch · Transformers · Multimodal ML
Hands-on cookbook of advanced RAG patterns — multi-query, fusion, re-ranking, hybrid retrieval — built on the OpenAI API. Stack: OpenAI API · LangChain · Vector DBs · Jupyter
Implementing "Attention Is All You Need" line-by-line — multi-head attention, positional encoding, encoder/decoder. No
nn.Transformer, just the math. Stack: PyTorch · NumPy · From-paper implementation
| Paper / Concept | Repository | Outcome |
|---|---|---|
| Machine Unlearning via SISA | AML_SISA_CODE_DEMO | PyTorch demo + YouTube walkthrough |
| Transformer (Vaswani et al.) | Transformer_From_scratch | Built from the paper — attention, encoder/decoder |
| Micrograd (Karpathy) | Micro_grad | Autograd engine, ground-up |
| PaliGemma (Vision-Language) | Vision-Language-Model | Multimodal architecture re-implementation |
| Quantum Neural Networks | Image_classification_by_QNN | QNN-based image classifier |
| Neural Collaborative Filtering | NCF_Recommendation | MIT-licensed reference impl |
| Advanced RAG techniques | RAG_Techniques_WITH_OPENAI | Hands-on cookbook |
| ProGAN — Progressive GAN | ProGANS_cv | Image generation pipeline |
🤖 LLM & Agents
🧠 ML / DL
📊 Data
🌐 Backend & Web
🗄️ Data Stores
☁️ Cloud & Tools
- 🧠 Mastering TensorBoard with PyTorch
- 🛰️ U-Net for Satellite Image Segmentation
- 📚 All You Need to Know About the "collections" Module in Python
Open to: Research collaborations · AI Engineering / Agent roles · Mentoring via Topmate
- 🤝 Open-source contributor to agno (agent framework) and OpenAI Cookbook
- 🌟 Stars across multiple repos — Solar_Rooftop_Detection, Vision-Language-Model, Transformer_From_scratch, NCF_Recommendation
- 🔬 Self-driven implementations of foundational papers — Transformer, Micrograd, SISA, PaliGemma
- 📚 Published Medium author — PyTorch, segmentation, Python internals
- 🎤 YouTube technical creator — SISA Machine Unlearning walkthrough
- 🎓 Pursuing M.Tech Machine Learning @ DA-IICT
"The hardest part of AI isn't the models — it's the wiring between them."