Hacker

Jeevesh Chaurasiya

@jeevesh2605

Joined Sep 2026Student at Pranveer Singh Institute of Technology, Kanpur Uttar PradeshIndia
  • BurnSwitch: Stop a Runaway AWS Bill with AI Agents and One Human Tap

    Flip the Switch HackathonNov 2026

    One bad deploy can quietly burn money all night. When a Lambda worker starts failing on SQS messages with no dead-letter queue, SQS retries the same batches endlessly. Costs climb, good orders get stuck alongside the bad ones, and nobody notices until the bill arrives. Fixing it means spotting the burn, finding the bug, pausing safely and recovering in the right order. Handing all of that to one autonomous agent means nobody signs off on commands that change production. BurnSwitch puts the whole incident in one Switch room bridged to Telegram. A custom watchdog agent prices the Lambda’s spend in real time and hands the incident to Claude Code exactly once. Claude Code reproduces the bug, writes a tested patch and proposes each command, but nothing reaches AWS until the on-call human taps Allow once on their phone. The watchdog verifies every change and counts every order, and a late-joining auditor agent writes the post-mortem. It’s for engineers running event-driven workloads on AWS who want AI help during incidents without giving up control.

  • Top 50

    RingBreaker: An AI Agent That Breaks UPI Mule Rings on FalkorDB

    Graph HacksOct 2026

    When someone in India is scammed over UPI, the money is usually gone within hours. It moves through a chain of mule accounts: new accounts with minimum KYC, each passing on amounts small enough to look normal. Fraud teams look at one alert or one transaction at a time, so every hop seems harmless. The pattern only appears when you follow the connections between accounts, devices, phone numbers and payments. By the time someone traces it by hand, the money has been cashed out. RingBreaker is for bank and payment-app fraud teams who handle victim complaints and suspicious-activity alerts. It stores accounts, UPI IDs, devices, phones and transactions as a graph in FalkorDB. When a complaint comes in, an AI agent investigates with multi-hop Cypher queries. It follows the money forward and backward, checks how quickly each account passes money on, and looks for shared devices and phone numbers. It then gives a structured verdict with the role of every account in the ring and evidence tied to real graph facts. Most tools stop at "flagged." RingBreaker goes three steps further. It recommends freezes, and a human must approve each one. It records every action in an audit trail in the graph. Then it re-queries the graph to prove that no money route from the victims to the cash-out account remains. It also leaves honest look-alikes alone, like a busy merchant or a family sharing one phone. All data is synthetic.

  • SmokeGuard: Estimating Smoke-Related Air Pollution Across Delhi-NCR

    Environmental HacksOct 2026

    https://builder.aws.com/content/3KXCpxUTlHjekaegysdjIFcjn35/smokeguard-estimating-smoke-related-air-pollution-across-delhi-ncr

  • VaporTrace: Real-Time Distributed Tracing for AWS Event Pipelines

    First CommitSep 2026

    https://builder.aws.com/content/3JZZVJmLVcQYvVrIJSopPEqOhhH/vaportrace-real-time-distributed-tracing-for-aws-event-pipelines