Each slip is one job a person did by hand: start a check, email a former manager, read the reply, chase the silence.
I'm a backend engineer in Bengaluru. I take the repetitive part of a team's day and turn it into a pipeline that starts, chases and closes work on its own, with Node.js, queues and LLMs.
White is waiting. Black is handled by the system. Blue is a person, in ink, on purpose: anything unsure still goes to someone.
Try one by hand: click a slip.
116 tasks a day, by hand
Checks that start, chase and close themselves.
Ops started every check by hand, emailed every former manager, read every reply and chased every silence. I built the pipeline that does it: queued outreach every 24 hours, an in-product form, a threaded inbox, and an LLM on AWS Bedrock that reads each reply and routes it. Unsure replies still go to a person.
- 116 to 14
- manual tasks a day. That is 102 tasks a day nobody does by hand.
- 86%
- of checks start with no human
- -31%
- turnaround, 22.4 to 15.4 days
Production data, 1-9 Sep vs 8-16 Oct 2026. Stack: Node.js, BullMQ, Redis, AWS SES, AWS Bedrock, React.
Your version: Replies pile up in a shared inbox? Tell me about it
Read a reply yourself
Arrives
Read
Routed
One check, start to finish
This toy reads your text with simple rules. The real pipeline uses an LLM, and the rule that matters is the same: when it is not sure, a person decides.
Day 1
- A check is createdsystem
- It becomes a jobsystem
- The referee gets an emailsystem
After 24 hours
- No reply in 24 hourssystem
When the reply lands
- The reply arrivessystem
- An LLM reads itsystem
Confident: it closes itselfUnsure: it goes to a person
What I can build for you
Replies pile up in a shared inbox
Queued follow-ups and an LLM that reads each reply and routes it; unsure ones go to a person.
Seen in Reference checksCustomers talk to you on WhatsApp
Cloud API, templates, webhooks and one inbox your team actually uses.
Seen in WhatsApp on Meta Cloud APISomeone checks documents by hand
Vision and LLM steps that fail safe to a person instead of guessing.
Seen in AI in the pipelineTwo screens have to stay in step
Realtime sync, presence and chat that hold up on bad networks.
Seen in PawseA small business needs its own app
Menus, bookings and admin panels the owner runs without calling a developer.
Seen in Bowl n Sticks
Not on the list? If a person on your team copies, chases or checks something by hand every day, it is probably a pipeline.
Tell me the processNine in ten replies sort themselves
Our own messaging pipeline instead of a reseller.
Outreach ran through a reseller with one polled status and no control over templates or replies. I proposed replacing it and built our own: send pipeline, webhooks, BullMQ queues, template sync, an agent inbox and a reply classifier, rolled out behind flags from 1% to 100%.
- 11K+
- messages in the first 10 weeks
- ~90%
- of replies routed with no agent
What went wrong: a 10% stage with no fallback stalled phone-only campaigns; fixed with a ramp and a reactivation script.
Messaging database, first 10 weeks. Here, 100 slips stand for all the replies.
Your version: Customers talk to you on WhatsApp? Tell me about it
Crooked photos, straightened
Vision and LLM steps that fail safe to a person.
About a quarter of uploaded documents needed someone to crop them by hand. A YOLO + OpenCV service now finds the document, straightens it and closes the ops task when it is sure. Free-text triage uses an LLM that sends anything ambiguous to a human.
- 2.25x
- recall, 19.5% to 43.8% correct on 411 production images
- Fails closed
- ambiguous cases still go to a person
Offline benchmark on 411 production images. One slip per image: 80 handled before, 180 now, the rest still go to a person.
Your version: Someone checks documents by hand? Tell me about it
One movie, two cities, in step
Pawse, live, open source.
“3, 2, 1, play” drifts apart. Pawse sends only tiny control events over a WebSocket room, corrects clock drift by nudging playback rate, and moves your own file peer to peer over WebRTC.
- 350 to 6 ms
- median gap between two players while playing
- 13 tests
- zero infrastructure cost
scripts/measure-sync.mjs, 150 ms simulated delay each way, 3 runs, 18 Oct 2026. Stack: React, Node ws relay, WebRTC.
Your version: Two screens have to stay in step? Tell me about it
Chat that tests its own locks
BuzzBell, live, open source.
Postgres row level security is the entire authorization layer; realtime presence, typing and messages ride on top. The interesting part is the tests that try to read what they should not.
Push on the wall with your cursor.
- 34
- offline SQL checks try to break the policies
- 12
- live RLS tests from real guest sessions
Stack: React, TypeScript, Supabase, Postgres RLS.
Your version: Two screens have to stay in step? Tell me about it
A cafe menu the owners run
Bowl n Sticks, live, freelance client.
A cafe in Karwar needed a menu guests open from a table QR, and owners who never wait on a developer. Permanent QR codes, drag-and-drop menu admin, privacy-safe analytics.
“Every menu change used to mean waiting on someone else. Pradumya built a menu our guests open straight from the table, and an admin we run ourselves: new dishes, prices and photos in a minute, from a phone. We haven't needed him for a single change since.”
- 141 items
- in 23 sections, managed by the owners
- 83 tests
- on the server actions
Read-only query of the live menu database, 19 Oct 2026. Stack: Next.js, Supabase, Postgres RLS, Tailwind.
Your version: A small business needs its own app? Tell me about it
A brochure you climb
Swastik Platinum, live, rebuilt study.
First built for a hiring assessment, then rebuilt: the page rides a 22-storey tower floor by floor and reads units straight off the real plans.
- 22 floors
- scroll-directed, floor and unit picker on the real plans
Stack: React, Tailwind, Vite.
All the work
Hiring for a backend role? Resume, PDF or LinkedIn.
- Reference checksProduction automationin production
- WhatsApp on Meta Cloud APIProduction automationin production, my proposal
- AI in the pipelineProduction automationin production
- PawseOwn productlive, open source
- BuzzBellOwn productlive, open source
- Bowl n SticksFreelance clientlive, freelance client
- Swastik PlatinumRebuilt studylive, rebuilt study
Who you would work with
I'm Pradumya, a backend engineer in Bengaluru. My job is finding the work a team still does by hand and building the pipeline that takes it over: queues that start and chase work, LLM steps that read what comes back, and a person in the loop for anything the system is unsure about.
Before this I built software for a US healthcare company. I also build my own products, like Pawse and BuzzBell, end to end.
“Out of the 100+ interns we've seen, Pradumya is that one person who stood out, not just with skills, but with sheer consistency, curiosity, and commitment. … I barely have to guide him anymore. He takes ownership, does his research, and always comes back with something that surprises us, in a good way.”
Jul 2026 to now
SDE-1, backend and AI automation
A background-verification platform
Joined as an intern in Jan 2026. 148 merged PRs across a Node.js backend, three React frontends and a Python AI service.
Jan to Aug 2025
Software Engineer Intern
GetMax Solutions, US healthcare revenue cycle
Built getmaxhealthcare.com end to end, and a multi-tenant workforce system with automated workflows and EHR API integration.
Aug to Nov 2025
Teaching Assistant, Design and Analysis of Algorithms
IIIT Vadodara
2022 to 2026
B.Tech, Computer Science and Engineering
IIIT Vadodara
CPI 8.79 / 10.
How we would work
A free 30-minute intro call
You walk me through the process your team does by hand. I tell you what I would automate first, and what I would leave to a person.
A written scope
What gets built, what does not, and how we will know it works, on one page before any code.
A small first milestone
One piece of the process running for real, so you see it work before committing to more.
Weekly demos on a live link
Every week you click through the current build yourself, no screenshots.
You own everything
Your repo from day one, with tests and handover notes, so anyone can run it after me.
I am in Bengaluru, India (IST, UTC+5:30). Taking projects now, 10-15 hours a week. Live overlap with US mornings and UK and EU afternoons; Asia by arrangement.
Fixed price per milestone, agreed in writing before any code. Book a free 30-minute call and you'll get a written quote after the call.
Book a free 30-minute callTaking projects now, 10-15 hours a week. Live overlap with US mornings and UK and EU afternoons; Asia by arrangement.
- Call
- cal.com/pradumya
- in/pradumya2004
- GitHub
- pradumya004
- Resume
- Based in
- Bengaluru, India (IST, UTC+5:30)