Jay. I build products end to end: the idea, the system under it, and the last 10% that makes it feel right. Agents write most of my code now. I make the calls. Lately that's SketchPen and Otto.20, based in Nagpur. AI engineer at Encraft, before that Prepairo. I care more about the problem than the stack, and I'm happiest where product design meets hard systems work.
Type a topic, get a hand-drawn explainer video: every line drawn stroke by stroke, narrated in English or Hindi. I wrote the drawing engine myself, with computer vision deciding what to draw and in what order, synced to the voice frame by frame. It took 30+ iterations to match the best in the category, at a better price. How I got there.
Run coding agents on a cloud machine and steer them from your laptop or phone. Paste a sandbox key and Otto sets up a Linux box, installs OpenCode, keeps it alive as a service and connects over HTTPS. Your agents keep working when your laptop sleeps.
Hybrid RAG for international enterprise organizations, where a wrong answer becomes a compliance problem: 100+ regulatory PDFs of 100+ pages each, eval harnesses that score relevance and citation grounding instead of yes/no, and pipelines that pick up where they crashed. Now building a multimodal video-analysis platform for a Fortune 500 global advertising network, scoring creator content with structured LLM evaluation.
Built the daily content engine: 100+ articles from 15+ sources in, 10,000+ GRE questions out, every day. Owned the GRE backend (FastAPI, async Postgres), built the AI tutor with memory and tool use, and ran the AWS deploys.
I'm not loyal to a language. Python, Rust, TypeScript: whatever solves it and stays boring in production.
I run several coding agents at once. They're fast at code. Architecture, trade-offs and taste stay with me.
Anything with a model in it gets evals: scores with reasons, deterministic checks, and real human review.
Something real in days. Then the last 10% that makes it feel right, because product design is part of the job.
Your coding agent in a Telegram DM. Streams like Telegram's own AI bots, with approvals and questions as tap-to-answer cards. No open ports.
Recurring agent jobs on systemd and launchd timers, with catch-up after sleep and results delivered to Telegram.
A layout DSL that lets LLMs draw audio-synced whiteboard lessons without ever touching a coordinate. On npm.
Idea to a full manga in about a minute, with characters that stay consistent from panel to panel.
Run coding agents on a cloud machine and steer them from your laptop or phone. Paste a sandbox key and Otto sets up a Linux box, installs OpenCode, keeps it alive as a service and connects over HTTPS. Your agents keep working when your laptop sleeps.
Your coding agent in a Telegram DM. Streams like Telegram's own AI bots, with approvals and questions as tap-to-answer cards. No open ports.
Recurring agent jobs on systemd and launchd timers, with catch-up after sleep and results delivered to Telegram.
A layout DSL that lets LLMs draw audio-synced whiteboard lessons without ever touching a coordinate. On npm.
Idea to a full manga in about a minute, with characters that stay consistent from panel to panel.