Built Mercury, an observability platform live at 1M+ subscriber ISPs, plus Root Minds, a multi-agent RCA system integrated with it and used daily by test teams.
Test and ops teams had no way to see subscriber- and device-level traffic during testing, so bugs slipped straight into production.
Built Mercury, a centralized observability platform (Fluentd/Loki logs, Thanos/Prometheus metrics, Grafana auto-mitigation) now running across 2 major ISPs and 4 internal test teams that together serve 1M+ subscribers.
Took Root Minds from a hackathon build (top 10 of ~100 at Cisco's Innovation Hackathon) to production: a multi-agent system that automates root-cause analysis by correlating logs, metrics, and alerts, then folded it into Mercury as a standing module, replacing the manual evidence-correlation an engineer used to do by hand.
Owned AutoResync, automating the full recovery process an operator used to run by hand after every network failover, control-plane resync, route resync, and session resync in one pass, validated at 128,000-subscriber scale and saving 10+ minutes of manual work per transition.
Architected a scheduled collection microservice end-to-end that runs live diagnostic commands on its own schedule, replacing the one-off scripts engineers used to run and the long-lived SSH sessions, which carried a 300-second timeout and MITM exposure.
Launched IPv6 support to meet a major Indian ISP's requirement, validated it at 1M+ subscriber scale, and hardened the service with live config reloads, storage health checks, and automatic circuit breakers so a failing dependency can't cascade.
Technical Intern · Cisco
Jan – Jun 2024
Automated subscriber redistribution across line cards; 16K subscribers rebalanced, zero downtime.
Hardware line cards were overloading past their subscriber limit, a placement problem later recognized by the Broadband Forum (BBF).
Automated the redistribution in Python to hold an 8K-per-card cap across 16,000 subscribers on 2 line cards, replacing a manual process with zero downtime.
Built a React Native + MongoDB dashboard so three network providers could see how subscribers were spread in real time, cutting operator investigation time 30%.
DevOps Intern · Reeudo
Jun – Jul 2023
Live FX-rate pipeline and client-side encryption for 1,000+ users.
The product needed live exchange rates and safe handling of financial data.
Built a real-time FX-rate pipeline (Google Apps Script to AWS S3) serving 1,000+ users on a platform processing live cross-border transactions.
Secured dashboard data for 400+ consultants using client-side AES encryption (CryptoJS) on AWS Lambda.
Set up Jenkins CI/CD for Flutter deploys to a 1,000+ customer base.
SDE Intern · MapmyIndia
Jun – Jul 2022
Halved a multi-modal routing API's error rate; +40% faster JSON parsing via a C++ nlohmann wrapper.
The multi-modal routing API (Valhalla engine), used by 5,000+ people to plan trips, was failing too often and re-deserializing every Transitland REST API response from scratch.
Slashed the error rate 50% and cut JSON parsing overhead 40% with a C++ nlohmann wrapper that eliminated repeated deserialization on every Transitland call.