Dibbayajyoti Royfull-stack engineer, Rust systems builder
in 10 seconds
Based in India · IST (UTC+5:30). Remote-friendly, EU relocation-eligible.
from a maintainer
“Thanks Dibbayajyoti Roy for an amazing contribution! 👏👏👏”
who
I'm a Full Stack Software Engineer and Rust developer at Yupcha Softwares Pvt. Ltd, where I architected and rebuilt a production AI Video Interviewer from the ground up: migrating a high-latency STT → LLM → TTS pipeline to a real-time speech-to-speech architecture on AWS Bedrock Nova Sonic, LiveKit, Pipecat, and WebRTC. In parallel I build open-source developer tools: Rust backend engineering, Rust + WebAssembly, and RAG / AI agent infrastructure. I also ship the HR SaaS modules around the interviewer: REST APIs, PostgreSQL, Next.js performance, and Linux infrastructure.
now
My focus is making the AI Video Interviewer — an agent that conducts candidate interviews end to end — better and ready for public use. Recent production work around it: cutting Redis polling latency by ~90% with batched MGET operations, resolving LiveKit and FFmpeg audio-video sync issues, and reducing page load time by 44% through N+1 query elimination, PostgreSQL indexing, and route splitting.
what I build
Three products anchor what I do publicly. Each solves a real problem I hit and could not find a clean answer to. The detailed writeups live on /work, and deeper topic pages cover Rust engineering, WebAssembly, AI engineering, Next.js SEO, and distributed systems.
AHTML: AI agents read websites slowly and expensively, so I built a sixteen-package npm scope (@ahtmljs/*) plus a Python SDK that lets any site publish an agent-readable, token-efficient snapshot, and emits MCP, OpenAPI, JSON-LD, llms.txt, RSL, and Markdown from one config.
Fresco: Windows has Wallpaper Engine and macOS has Lively, but every Linux option was terminal-only or abandoned, so I built a proper GTK4 desktop app that sets any video, GIF, or image as a hardware-accelerated live wallpaper: install a .deb, pick media, done.
whatbroke: When a test or server dies in my terminal I kept losing the context (what changed since it last worked, which file is to blame), so I built a zero-config, local-only CLI and read-only MCP server that wraps the command, writes a redacted, git-anchored crash bundle, ranks the suspect files with no LLM, and then verifies the agent's fix by re-running the exact captured command.
Diffcore: I needed a JSON diff fast enough for large client-server state sync and found the popular libraries either slow or unmaintained, so I wrote the engine in Rust and ship it to npm as WebAssembly.
Roy UI: Every dashboard I built started by re-wiring TanStack Table or copy-pasting shadcn and bolting on filters, so I built one React component library that ships a batteries-included DataTable, a no-dependency date range picker, and an analog/digital time picker, installed, not copy-pasted, with zero config.
Klinder-OSS: I kept stitching PostHog, LogRocket, and Customer.io together on every project, so I built one SDK that unifies event tracking, error-based session recording, and email trigger workflows in a single install.
how I think
I optimize for shipping. Production beats perfect. I prefer boring infrastructure (Postgres, nginx, systemd, Docker) and experimental edges (Rust, WASM, Cloudflare Workers, AI agents). I write before I commit and benchmark before I ship.
background
- 2024award
1st Runner-Up, NITA-ISRO Space Hackathon
- 2025award
Winner, NITA Arjuna 2.0 National Hackathon
200+ teams
- 2025award
Winner, Technovate Project Exhibition
- July 2025rolenow
Joined Yupcha Softwares Pvt. Ltd
Full-time, core team
- 2026award
Top 500, AI for Bharat Hackathon
- 2026education
Graduating, B.Tech in Computer Science
ICFAI University Tripura
Eligible for Germany Opportunity Card / EU Blue Card and the Netherlands Highly Skilled Migrant programme.
If you got this far, we should probably talk.
Book a call