Jagrit Pareek

• Lead product and engineering for PeopleSense HeatMap (formerly PeopleSense Places, since 2024), a live map of sensor occupancy and earthquake and public-safety alerts used at BITSync 2024, a 49ers ACE event train, OneLA and the Great ShakeOut. Own a 100K+ line codebase (330+ commits) and a 17-milestone roadmap inside a hard 40-60 USD/month budget.

Platform and security

• Replaced direct browser access to DynamoDB with an authenticated TypeScript API (Fastify, Kysely, Zod) on Postgres/PostGIS, removing AWS keys from the public bundle; a CI check fails the build if one returns.
• Delivered a portable single-server platform (Docker Compose, Caddy, Postgres) on one arm64 EC2 instance, provisioned with OpenTofu and deployed by GitHub Actions over OIDC with no stored cloud secrets; rejected Fargate/ALB/RDS (about 330 USD/month).
• Added an admin status page with rule-based email and Telegram alerting (stale feeds, silent earthquake stream, disk, failed backup), plus security headers, a report-only CSP and dependency-audit gating.

Real-time alerting

• Moved alert ingestion into one worker: 36 MeteoAlarm feeds, 5 national CAP feeds, FEMA IPAWS (signed SNS endpoint plus OpenFEMA reconciliation) and India's NDMA SACHET, with idempotent inserts, backoff and a one-year S3 archive.
• Built the multi-tenant MQTT alert broker (Mosquitto, ACL-scoped topics) so each agency receives only its own coverage area.

Earthquake early warning

• Built a Rust service for USGS ShakeAlert: persistent QoS 1 MQTT over verified TLS, one never-deleted record per earthquake keyed by USGS core id, and a fast path that pushes each message beside the database write with a per-version latency record.
• Wrote eew-replay and a CI suite that replays 11 recorded USGS earthquakes through Mosquitto into Postgres.

How it ships

• Shipped 50 pull requests in under four weeks, each with tests and CI gates, working with AI coding agents under a written testing and review standard.

India delivery, partners and people

• Ran India delivery for DMRC, Mumbai Metro and NCRTC (5 Delhi and 2 Mumbai site trips); presented the calibration pipeline to DMRC's GM Telecom, CISF Chief Security Commissioner and Director of Operations (POC and MoU approvals); delivered India's first permanent PeopleSense micro-cell at Rajiv Chowk (100% uptime after install).
• Built that site's calibration pipeline (a daily Lambda computing EWMA-blended profiles from fare-gate data) and ML on the same transit data (LightGBM, XGBoost, OpenVINO; 27% better inference accuracy).
• Drafted and finalised the FEMA IPAWS-OPEN MOA; wrote the WIS 2.0 whitepaper for UNDP EW4All Pillar 4 and CAP-XML research for NDMA's Sachet app; work with IAEM, OASIS Open, UNDP, USGS and EMTC on emergency relief and occupancy; invited to the 2026 CAP Implementation Workshop, UK (hosted by the Environment Agency, Alert-Hub.Org and OASIS Open).
• Mentored 5 student cohorts (FCUSD, BVT, BITS Pilani PS1, Sacramento State) on Python, Raspberry Pi, MQTT, AWS and AI coding.

• Took PeopleSense Monitor from a solo practice-school prototype to a multi-tenant production dashboard for 11+ client organisations (ACE Rails, CCJPA, USGS, Sac State): live occupancy for 5+ rail corridors, a digital-twin station view, admin roles and SES notifications.
• Owned a multi-account AWS estate (35+ Lambda functions, 50+ DynamoDB schemas, EC2, Amplify CI/CD, Route 53, IAM); a billing audit cut unnecessary Lambda and DynamoDB spend.
• Rebuilt sensor auto-deployment for the Greengrass V1 to V2 migration (pi-gen OS images, hotspot web-form onboarding, LED status), removing manual SSH from every install; deployed and maintained 100+ IoT sensors across train cars, schools and Delhi Metro.
• Built the company's first mobile app, PeopleSense-BLE-Jig (React Native), for Bluetooth Classic and BLE scanning with RSSI history and geolocated PDF export; also wrote a native Kotlin BLE scanner with AI help.
• Wrote technical sections for 4 RFPs (DFW Airport, AMTRAK, Sourcewell, CCTA), all submitted.

• Researched lead-free thermal-resistance (stabiliser) additives for PVC pipes; mapped plant workflows and cut identified process inefficiencies by 15%.