Sach
A WhatsApp forward, checked in seconds - matched against PIB Fact Check's archive first, then a live AI check grounded in official sources.
Senior GCP DevOps Engineer at UKG. I build cloud platforms for a living, and AI agent systems that have to show their work as proof of craft.
Drag to rotate, tap a node to inspect it - or scroll on for the guided tour.
Live WhatsApp-forward fact-checker. Paste a forward, get a cited TRUE / FALSE / MISLEADING / UNVERIFIED verdict.
I am a Senior GCP DevOps Engineer at UKG, building cloud platform infrastructure since 2018 - currently on the UTM infrastructure team, working on GKE platforms, a SQL Server 2022 migration, and PagerDuty noise reduction. Based between Hyderabad and Toronto, Canadian permanent resident.
In parallel I design and ship AI agent systems end to end, solo: a public-interest fact-checker used from WhatsApp, an autonomous options-research agent, and a live paper-trading desk. One rule runs through all of them: every claim an agent makes must be traceable to evidence, and "I could not verify this" is a valid answer.
Paste a viral WhatsApp forward and get a cited TRUE / FALSE / MISLEADING / UNVERIFIED verdict from PIB Fact Check and official sources, in English and 15 Indian languages, as a card ready to post back into the family group. Built solo, runs on free tiers, filed as submission #126.
A viral forward reaches a family group in seconds; a human fact-check takes days. By then it has been believed, forwarded and acted on.
A three-layer cascade: match the claim against PIB Fact Check's archive and published fact-checks first, then a live AI check grounded in official sources, and if the evidence is not there, an honest UNVERIFIED with a human tipline handoff - never a guess dressed as a verdict.
Public 79-claim evaluation with zero overclaimed FALSE verdicts, abstention preferred over guessing. Live at sachcheck.in in English and 15 Indian languages, filed in the Code for a Billion judging window.
79 claims - 0 overclaimed FALSE - live at sachcheck.in
An autonomous pre-market options research agent for US markets. It investigates 12 names across fragmented sources and produces trade cards where every number traces to the evidence packet it came from - including a verifier that catches the model misquoting its own data. Screening rank 252 of 547; demo score 12.5 of 15.
Research agents sound certain while misquoting their own data - and a trade idea built on a wrong number is worse than no idea at all.
A multi-agent pipeline investigates 12 names across fragmented sources and assembles evidence packets. Every number on a trade card must trace back to its packet, and a separate verifier pass exists to catch the model misquoting its own evidence.
Open Agent Hackathon 2026: screening rank 252 of 547, with a demo score of 12.5 of 15 - higher than all three Track 02 winners' demos (12.3 each).
Rank 252 of 547 - demo 12.5 of 15, above all three Track 02 winners
A Nasdaq-100 options research desk running on virtual money: technical setups, an intraday news radar, market-regime filters, and strict risk rules with a drawdown kill-switch. The public page refreshes every 5 minutes in session, with an interactive 3D diagram of the whole system.
Strategy ideas are cheap. The hard part is a desk that sizes positions, takes profit, stops out and flattens by rule - proven without risking real capital.
Nasdaq-100 technical setups pass through an intraday news radar and market-regime filters into a risk engine with conviction sizing, take-profit and stop-loss rules and a drawdown kill-switch. Everything flattens at the close; paper money only, by design.
Runs in public as a live desk: refreshes every 5 minutes in session and republishes itself daily, with every position and exit on the page.
Live desk - 5-minute refresh in session - 0 real orders
On desktop, keep scrolling - the highlights travel sideways
Three tiny canvas toys - the kind of play that keeps the serious work sharp. All three run on one shared animation loop and pause off-screen.
UTM infrastructure team: GKE and Cloud Run platforms, Terraform/OpenTofu with Terragrunt, Argo CD GitOps, SQL Server 2022 migration, PagerDuty noise reduction, FinOps with Kubecost and Spot VMs.
Cloud and CI/CD engineering in a regulated banking environment.
Release and infrastructure engineering for enterprise software.
Linux systems administration - where the infrastructure career started.
One-year postgraduate diploma, completed while working full-time in DevOps.
One-year postgraduate diploma, completed alongside the George Brown diploma.
Bachelor of Technology - the engineering foundation before the move into infrastructure.
Build notes from the two solo systems - the honest versions, numbers included.
Why I built a fact-checker that has to cite its verdicts, and what its 79-claim evaluation actually showed. Read on dev.to - the launch post on LinkedIn.
The QuantPilot build story from the Open Agent Hackathon: a demo that outscored all three track winners, and the innovation gap that kept it out of the finals. Read on dev.to - the original on LinkedIn.
Certifications: Google Cloud Professional Cloud DevOps Engineer - PagerDuty Foundational Practitioner.
Education: BTech, CVR College of Engineering, Hyderabad - PG diplomas in Cloud Computing (George Brown) and Business Analytics (Cambrian).
Toronto hybrid or Canada remote. The fastest way to evaluate my work is the live products above - Sach takes about 10 seconds to try. For roles or collaboration, LinkedIn is the best channel.
For roles and collaboration - the fastest response.
Connect →Every project public - code, evaluations and all.
View profile →Try the work directly - a verdict in about 10 seconds.
Try it →84 badges - certifications, publicly verifiable.
Verify →The build stories, written up in full.
Read →Builds and demos on video.
Watch →