Portrait of Atul Chahar

Atul Chahar

AI engineer · Bengaluru, India · founder of EliteFolks

Portfolio · GitHub · LinkedIn · X · LeetCode · Codeforces

Who is Atul Chahar?

Atul Chahar is an AI engineer and B.Tech student in Bengaluru, India, and the founder of EliteFolks. He builds AI agents, on-device AI and the checks that keep them honest: a model proposes, code verifies, a person approves.

I’m a second-year B.Tech student in Bengaluru who builds agents, automations and LLM apps. AI writes most of my code. I spend my own time on the design, the checks, and the parts that break. Kavach took 36 hours. KILLSWITCH took four days. B.Tech in Computer Science (AI and ML) at Medhavi Skills University with Polaris School of Technology, 2025 to 2029. I founded EliteFolks in 2026. I work in Claude Code every day, with Next.js, TypeScript, Python and Kotlin underneath. Open to a paid internship or a full-time role.

Quick facts

Name
Atul Chahar
Role
AI engineer, B.Tech student, founder of EliteFolks
Location
Bengaluru, Karnataka, India
Education
B.Tech in Computer Science (AI and ML), Medhavi Skills University with Polaris School of Technology, Bengaluru, 2025 to 2029
Focus
AI agents, on-device AI, verifying model output in plain code, human approval for irreversible steps
Stack
Claude Code daily; Next.js, TypeScript, Python, Kotlin, AWS
Open to
A paid internship or a full-time role

Awards and results

Selected projects

Outspoken Accessibility app, 2026

People who can’t move or speak talk out loud by blinking at a phone. It works without the internet.

A patient on a ventilator or someone with ALS is often awake and clear-headed but can’t say “I am in pain”. Outspoken reads their eyes through the front camera, lets them pick a reply or spell out anything, and speaks it in a natural voice, all on the phone. My 14 merged pull requests cover the designed screens, 17 languages, the patient profile, switching between Hindi and English, a Kokoro voice and safer model loading.

Role: Contributor, 14 merged pull requests · Team project. Semi-finalist, Tata Imagination Challenge 2026 · Kotlin · On-device Gemma · Kokoro TTS

DENO Multi-agent system, 2026

Eight agents argue over a startup idea and give you a verdict you can defend.

Most AI tools will happily build a landing page for a bad idea. DENO asks first whether the idea should exist. Research, critic, strategy and decision agents challenge each other: they size the market from the bottom up, find competitors with sources you can click, and map the communities where the first users are. Only an idea that survives moves on to a 30-day content plan written in the founder’s voice, with every post scored for ban risk.

Role: Agent design, with Krishiv Agarwal and Siddhartha Singh · 2nd place, Lyzr Agentathon · Multi-agent orchestration · LLMs · TypeScript

Kavach On-device scam detection, 2026

Spots scam calls on the phone itself, in Hindi, Hinglish and English. Built in 36 hours, with no internet permission at all.

The pitch opened with a real call: a stranger said my father had sent money, then a fake bank alert for ₹30,000 arrived, then a request to return ₹27,000. Kavach listens for those patterns during a call and warns you while it’s still happening. Everything runs on the phone, and the internet permission is stripped from the manifest so anyone can check. One script re-measures all 31 claims in the README.

Role: Android and on-device LLM, Team Smoke Test · Second runner-up, iQOO Hackathon 2026, Bengaluru · Kotlin · On-device LLM · 136 tests

KILLSWITCH Cloud security, 2026

Contains leaked AWS keys. A model proposes, code verifies, an operator approves. Built in four days.

Push an AWS key to GitHub by mistake and bots find it in about five minutes. AWS quarantines the key, but it doesn’t shut down what the attacker already started. KILLSWITCH catches the leak, reads CloudTrail across regions to see what the key did, has a sandboxed model propose a containment plan, checks every line of that plan against the audit facts in plain Python, and waits for an operator before it touches anything. I designed the architecture. Gyan Ranjan built the approval console and Krishna Chaudhary the attacker script.

Role: Architecture, team of three · First Commit, WeMakeDevs and AWS · Python · AWS CDK · React · 404 tests

Foreman Agent harness, 2026

A supervised swarm of coding agents works through a backlog in parallel. Anything irreversible waits for a human.

Foreman hands issues to a bounded swarm of AI implementers, each on its own branch inside a sandbox, then merges their work through a serial queue with a reviewer stage. A risk-tier policy stops anything irreversible at a human gate. Auto-approving a dangerous step is a compile-time false, not a setting someone can flip. State lives in SQLite, so a crashed run picks up where it stopped. Built on TrueFoundry’s open-source TrueForge harness.

Role: Design and build · Agent Harness Hackathon, WeMakeDevs and TrueFoundry · JavaScript · TrueForge · MCP · SQLite · 118 tests

EliteFolks Learning platform, 2026

Coding practice that picks your next problem from how you did on the last ones, with a compiler for 80+ languages and a voice tutor.

I started EliteFolks so students would stop solving random problems. It runs code in the browser in more than 80 languages, and has a real-time voice tutor, 1v1 coding battles and an AI interview simulator. About 1,100 people signed up and 417 were active, mostly from my college. The first beta had more than 250 users, and it launched on Product Hunt.

Role: Founder · Since February 2026. About 1,100 sign-ups, 417 active · Next.js · TypeScript · Appwrite · AWS · n8n · Gemini

Kaagaz Graph and agents, 2026

Tells a student whether they’re already too late for a scholarship, with a government source for every rule.

Scheme finders tell a student what they’re eligible for. Kaagaz tells them whether they’re already too late. It walks the chain of certificates a scholarship needs in a FalkorDB graph, adds up the legal processing time for each, and names the document to apply for today. A caseworker agent acts only after approval, and a watcher agent re-checks the rules against the live government pages.

Role: Design and build · Graph Hacks, WeMakeDevs and FalkorDB · FalkorDB · Cypher · Agents

AEGIS Safety platform, 2026

Tourist safety for trails with no signal and no power: an offline mesh, a command map and a digital ID.

Built for places like Meghalaya’s canyons and rainforests. Phones pass incidents to each other over an offline mesh, an authority dashboard shows breadcrumb trails and search heatmaps on a map, and a privacy-preserving digital ID lives on a blockchain.

Role: App, backend, dashboard and contracts · Kotlin · Node · React · Leaflet · Solidity · 48 tests

Also built

Open source

Hackathons and events

Right now

Contact

Reach Atul on GitHub · LinkedIn · X · LeetCode · Codeforces.

Last updated 2026-11-09. A plain-text version for AI agents is at /llms-full.txt.