I'm Gobinda, a senior software engineer who designs, builds and ships products end to end, now with AI at the core.
years shipping products, frontend to backend
faster repo read at worker startup, 278.7 s to 9.5 s
Web Vitals score, up from under 10 at BitClass
fewer API calls after reworking state and fetching
At AptEdge I don't stop at the edges of a ticket. I talk to customers, shape what we build, design it, build it, and look after it in production.
On calls with customers: hearing the problem first-hand, debugging their issues, and closing the loop when the fix ships.
I redesigned and rebuilt AptEdge's product: the AI search, the answers, the integrations and the way in. Drag across each screen to compare the old app with the new one.
AI search across every connected source: one question, answered from tickets, articles, edges and docs.
It reads closed tickets, checks each one against the articles and docs that exist, and clusters what's left into recurring issues. One article per cluster can close dozens of tickets at once.
By day I work at AptEdge, which builds AI agents for B2B technical support. On my own time I build Glance, a voice explainer for Claude Code. Here's its pipeline.
Cloud voices (ElevenLabs, MiniMax) slot in at the voice stage, fetched one paragraph ahead, with the on-device voice as the fallback.
Glance follows live Claude Code sessions, renders each streamed reply as it lands, and lets you ask Claude about a selection right in the editor.
A headless Claude Code runs in auto mode when no terminal session exists. Next: drafting pull requests and fixing review comments with Claude.
A neural voice (Kokoro-82M through Core ML) speaks every reply offline, with ElevenLabs or MiniMax as a cloud option that fetches one paragraph ahead.
Specs, plans, reviews and parallel agents are part of how I ship, with the judgment to know when the model is wrong.
Case study
Our search workers had been restarting every few minutes for five years, and each restart took about five minutes. On a busy day they spent 73% of their time starting up, and pods were killed when both workers were down at once. I found two causes and fixed both.
def pre_request(worker, req):if req.path == HEALTH_CHECK_PATH:worker.nr -= 1 # probes don't age the worker
# before: 3,300 × git show <branch>:<file>git ls-tree -r <branch>git cat-file --batch # every blob, one process
A voice and visual explainer for Claude Code: replies rendered beside the terminal, read aloud, diagrams drawn as it speaks.
Review, manage and ship pull requests without leaving Slack. Approvals, changes and merges sync both ways with GitHub; reply, assign reviewers and approve right in the thread.
The new site for AptEdge's AI support engineer, designed and built in Next.js: a live multi-hop investigation demo and a 3D knowledge-graph hero.
Rebuilt a slow React app as typed Next.js, then chased every render and request. About 40% fewer API calls along the way.
A grocery delivery app for local shops. I built it, shipped it, onboarded the vendors, and sold it to a local business owner.
Point of sale and stock systems for businesses across retail and hospitality, plus a school website I ran for three years.
AptEdge builds AI agents for B2B technical support. I own features end to end: talking with customers, shaping the product, designing the UI, and building it front to back.
Took a slow React app to a fast, typed Next.js codebase.
Built a grocery delivery app for local businesses, from first commit to exit.
Ran engineering teams delivering software for clients across retail and hospitality.
Deep in the frontend, comfortable across the stack, and the one who reviews the pull request.
Speaks English, Hindi, Bengali and Assamese.
Open to senior frontend and full-stack roles, and to the occasional product worth building. Bangalore, India.