Mohan Ranganath Varma Vummadisetty

Problem — Database migration is something I do professionally — and the tooling around it always felt heavier and less friendly than it should be.

Work — Building a desktop application that makes migrating databases into Snowflake approachable: an intuitive interface over the messy workflow of moving schemas and data.

Outcome — Combines everything I enjoy: database engineering, cloud, UI design, and workflow automation — in one tool I actually want to use.

Problem — Two people, one movie, different places. Existing tools made shared watching clunky.

Work — Built a real-time watch party app where playback stays synchronized while both webcams stream — exploring WebRTC, state synchronization, and real-time UX.

Outcome — A working app, and a crash course in the hard parts of real-time communication on the web.

Problem — Repetitive tasks accumulate silently — in development workflows, file handling, and daily routines.

Work — A growing collection of small utilities and workflows that automate repetitive work: dev-environment helpers, file processors, and glue between tools that don't talk to each other.

Outcome — Automation became one of my favorite areas of engineering — most of these tools save minutes daily, forever.

Problem — Cloud assistants don't belong in every corner of a personal machine. I wanted a persistent assistant that runs fully locally — inside the desktop environment I'd already built.

Work — Building Raphael into my Caelestia (Quickshell) setup on Hyprland: a QML drawer for the face, a FastAPI brain streaming from a local model via Ollama, one milestone at a time — chat first, then memory on Postgres + pgvector, then approval-gated actions.

Outcome — The first vertical slice runs: one keybind slides the drawer in and replies stream from a fully local model, themed to match the rice. Memory and safe actions are next on the roadmap.

Problem — A resume tells you what I've done. It can't show you what I love building — experiences where engineering and design meet.

Work — Designed and built this interactive 3D world: Next.js and React Three Fiber, GSAP scroll choreography, custom shaders for the data pipelines, and AI-generated hero assets compressed from 380MB of raw output into a world that loads in seconds.

Outcome — You're inside it. If it made you scroll one zone further than a PDF would have — it worked.

Problem — AI tooling moves fast, and reading about it teaches you almost nothing compared to building with it.

Work — Continuous experiments with AI agents, Claude Code workflows, local LLMs, MCP servers, prompt engineering, and ComfyUI pipelines — integrating them into real engineering workflows.

Outcome — A practical understanding of how modern AI systems actually behave — and a toolbox of working integrations.