I build agentic AI systems and data pipelines by day, and hack on side-projects — from MCP tools to a from-scratch OS kernel — by night. Currently building GenAI at IBM.
Software Engineer with 2+ years of experience designing scalable data pipelines, distributed cloud services, and full-stack web apps. I turn messy data into clean insights and sprinkle some AI magic on top.
I've been recognized for building agentic AI solutions and leading hackathon-winning projects. When I'm not engineering ETL workflows with PySpark, I'm probably breaking (and fixing) something in production.
Reduced memory footprint 70% by serving quantized Mistral 7B on FastAPI. Streaming responses in <2s latency with Tavily API for contextual search.
An OKLCH color-palette tool paired with an MCP server. Generates palettes from an image or hex, then pipes them into Claude Code, Cursor, or any MCP agent as Tailwind v4 tokens, CSS variables, or design-token JSON.
Automated creative writing across 5 modular agents — Genre, Character, Plot, Dialogue, Review. Improved coherence of generated stories by ~25%.
Conversational Pokemon metadata search by integrating Claude.ai with real-time PokeAPI calls using the MCP SDK. Gotta catch 'em all... programmatically.
A 32-bit x86 OS kernel built from scratch in C and assembly. Boots via Multiboot under QEMU with interrupt handling (GDT, IDT, PIC remap, IRQ-driven timer + keyboard) and memory management — a bitmap physical frame allocator and 4 MiB paging.
Won 1st place in a 24-hour designathon at NIT Trichy. Converted complex Figma designs into pixel-perfect, responsive React code under extreme time pressure.
Got a wild idea, a data problem, or just want to nerd out about LLMs? My inbox is always open.
ankitgdes@gmail.com