ProjectsPhotography AboutContact ResumeHome

Things I've shipped
and run.

Final-year CS (AI/ML) student working towards DevOps and SRE roles: a released desktop app, CI/CD pipelines, and load-tested, monitored services on AWS. Source for all of it is on GitHub.

TypeScript Released · auto-updating

StrataField

Installer cut from 134 MB to 6.4 MB · in daily use at a drilling business

Offline-first Windows desktop app used day to day by my family's drilling business to record borewells, soil layers, pipes, water levels and site photos, with an offline city map of water depths and geological cross-sections.

StrataField's home screen: water level over the years, the newest borewells and what needs attention
A borewell's page: details, soil layers, pipe and water level The map, coloured by depth to water A cross-section between borewells: measured columns with estimates between them
Screenshots from the released app.

Implementation

Rebuilt the original Electron version on Tauri 2, cutting the installer from 134 MB to 6.4 MB, and replaced sql.js with native SQLite (schema migrations, edit history, recycle bin, daily backups) that imports the old app's data automatically. GitHub Actions runs typecheck, lint, unit tests, a Windows end-to-end test and a clean-machine installer test; tagged releases publish signed builds that installed copies auto-update from.

Tech Stack

Tauri 2, React, TypeScript, Rust, SQLite, Leaflet, Vitest, GitHub Actions

Go AWS EC2

LoadLab

Go behind Nginx on EC2 · Prometheus and Grafana · three k6 load profiles

A load-testing and observability sandbox built to fail on purpose: Nginx in front of a Go backend with deliberately CPU- and memory-heavy routes, scraped by Prometheus and visualised in Grafana.

Implementation

Deployed as a Docker Compose stack on EC2, with the backend rewritten from Node.js to Go. Low, medium and high k6 load profiles show where p95 latency climbs and which resource saturates first. Fixed a Grafana redirect loop behind the Nginx sub-path proxy by correcting the root URL and forwarded headers.

Tech Stack

Go, Nginx, Docker Compose, Prometheus, Node Exporter, Grafana, k6, AWS EC2

Source ↗
Python Deloitte capstone · team project

IntelliOps

My part: CI/CD workflows, Kafka event bus, Helm chart

Deloitte capstone: an agentic AIOps platform for alert correlation, root-cause analysis and human-approved Kubernetes remediation. I contributed the CI/CD workflows, the Kafka event bus and the Helm chart.

IntelliOps overview: the live pipeline from detect to verify, noise reduction, and Prometheus signals for CPU, memory, latency and error rate
An incident collapsed from several alerts, moving through correlation, root-cause analysis, approval and verification Governance: RBAC, reversible-only and human-in-the-loop gates with pass and block counts The audit log of every approval, execution and escalation
Screenshots from the running platform: overview, an incident, governance gates and the audit log.

My Contribution

CI/CD workflows covering lint, tests, a Docker Compose smoke test, a frontend build gate and GHCR image builds. Added Kafka as an event bus interchangeable with the in-memory backend (checked by contract tests), a Helm chart for Postgres, Redis, all services and a migration job, and a load-test script.

Tech Stack

Python, GitHub Actions, Kafka, Helm, Kubernetes, PostgreSQL, Redis, GHCR

Source ↗
Python IEEE · under review

Invisible Image Watermarking (DWT + CNN)

Deep learning-based invisible image watermarking using the Discrete Wavelet Transform and a CNN. Written up as a paper with M. Talwar and S. Singh, currently under review at IEEE.

Evaluation

Measured with PSNR, SSIM and bit error rate under noise, JPEG compression, crop, scale and rotation attacks.

Tech Stack

Python, TensorFlow, Keras

Source ↗
TypeScript Web predecessor of StrataField

StrataDesk

Full-stack React/TypeScript app for borewell data (Excel import, map, cross-sections) with CI/CD via GitHub Actions, GHCR and AWS Systems Manager to EC2, plus Prometheus/Grafana monitoring and k6 load tests.

Python

Handwritten Math Expressions with Deep Learning

CRNN + CTC model in TensorFlow that reads handwritten math expressions from images and solves them, trained on Kaggle plus synthetic data.

LoadLab · 24 Jun 2026

Fixing a Grafana redirect loop behind an Nginx sub-path

One trailing path in proxy_pass sent Grafana and Nginx round in circles, and the rule that fixes it.

StrataField · 30 Sep 2026

134 MB to 6.4 MB: moving from Electron to Tauri

What made the old installer so big, what moved into Rust, and how existing data came along untouched.