Backend / AI Engineer
TypeScript
,React
, andNext.js
writing clean APIs withExpress.js
, managing data inPostgreSQL
, and running everything onBun
orNode.js
. I ship fast, debug faster, and keep optimizing until it just feels right.Places that trusted me with their codebase (and didn't regret it).
• Collaborated with a team of developers to build Snap-form, an open-source, AI-powered form builder designed to streamline dynamic form creation.
• Developed AI-powered form generation using LLMs to parse natural language prompts and dynamically generate JSON schema-backed React forms.
• Designed a PostgreSQL schema with Prisma ORM to store dynamic JSON form layouts, user submissions, and reusable templates.
• Integrated BetterAuth for OAuth 2.0 authentication and developed a background synchronization worker to stream submissions to Google Sheets.
• Architected the platform as a Turborepo monorepo with Bun workspaces, featuring a Next.js frontend, Express.js backend, and a dedicated cron service.
Interesting
A production-grade prototype of a multi-agent AI system for handling end-to-end customer support for CloudDash — a cloud infrastructure monitoring SaaS.
A production-grade prototype of a multi-agent AI system for handling end-to-end customer support for CloudDash — a cloud infrastructure monitoring SaaS.
A real-time collaborative coding environment with live code sync, AI assistance (Gemini), and multi-language support from JavaScript to Go.
A real-time collaborative coding environment with live code sync, AI assistance (Gemini), and multi-language support from JavaScript to Go.
Multi-provider LLM orchestrator built for a school-grade question answering app. Routes student queries to the cheapest model, rotates API keys automatically, and simulates 1,000 concurrent users.
Multi-provider LLM orchestrator built for a school-grade question answering app. Routes student queries to the cheapest model, rotates API keys automatically, and simulates 1,000 concurrent users.
A Jev-powered draft classifier that vibe-checks social posts in one call. Tone (Choice), viral potential (Score) and cringe risk (Noul) return as typed probabilities with confidence, rendered as bars plus a stamp verdict, no chat parsing.
A Jev-powered draft classifier that vibe-checks social posts in one call. Tone (Choice), viral potential (Score) and cringe risk (Noul) return as typed probabilities with confidence, rendered as bars plus a stamp verdict, no chat parsing.
Fetching your GitHub activity data
Hot takes, and things i figured out the hard way.
What TypeSafe's Jev System One model actually is, Choice, Score and Noul decisions with probabilities instead of tokens, and what I learned testing it on a school routing classifier and a tiny Postmark demo.
How I turned any JPG into a 3024x4032 Ray-Ban Meta frame with pure browser Canvas, EXIF surgery and a little gyro math, and what building a 100% local, Instagram Story ready converter taught me about images, privacy and shipping tiny tools.
midnight ideas and real talk that didn't fit into a tutorial.
Docker is used to to create and package application into containers. Kubernetes is used to orchestrate those containers…
Concurrency > Say we have 3–4 tasks/operations. When CPU executes 1 task and in most backend cases tasks are mostly I/O …
So, I was today year old when I found out that WebSocket connections at first are just simple HTTPS requests. The brow…
proof that i exist in three dimensions and have a personality.