Final year B.Tech CSE (AI/ML) student at Manipal University Jaipur, Class of 2027. Based in Lucknow, India.
Building toward DevOps / AIOps roles, with projects spanning observability, load testing, agentic incident response, and applied AI. Also building Strata, a desktop ecosystem for borewell and materials data.
Strata — a desktop ecosystem for digitizing and managing borewell and materials data, born out of my father's drilling business.
- StrataField — the core offline-first Windows desktop app. Tauri 2 / Rust / React / TypeScript / SQLite, with custom Excel ingestion, anomaly detection, lithology classification, an offline city map of water depths, and borewell cross-sections. Migrated from Electron, cutting the installer from 134 MB to 3.5 MB.
- StrataDesk — the web predecessor. Full-stack React/TypeScript + Express/PostgreSQL app with Excel import, a map dashboard and a cross-section engine. CI builds and smoke-tests the full Docker stack, pushes images to GHCR, and deploys to EC2 via AWS SSM; monitored with Prometheus/Grafana and load-tested with k6.
- StrataFlow Viz (private, in progress) — the visualization track for turning digitized strata records into visual geological insights.
IntelliOps CoE — an agentic AIOps platform for automated incident detection, diagnosis, and governed remediation. The Docker Compose stack includes telemetry ingestion, alert correlation, ML-based RCA, a human-in-the-loop approval gate, feedback-driven learning, a read-model service, and a React operator console. Approved fixes run as safe, reversible Kubernetes actions (scale, restart, rollback) on a local kind cluster, with a post-fix health check and an audit trail. Built as a team project.
Horizon Autoscaler (work in progress) — a Kubernetes-native predictive autoscaler that uses historical metrics to scale workloads ahead of traffic spikes.
- LoadLab — a Dockerized load-testing sandbox: Nginx and a Go backend with intentionally expensive endpoints, Prometheus/Grafana telemetry and alerting, k6 load generation, deployed on AWS EC2.
- Watermarking CNN — deep learning-based invisible image watermarking that combines the Discrete Wavelet Transform (DWT) with a convolutional neural network.
- Embeds a watermark in the wavelet domain so it stays imperceptible in the host image, with a CNN handling the embedding/extraction.
- Written up as an IEEE-format paper, co-authored with Manvik Talwar and Srajal Singh (MUJ), and taken through multiple rounds of review and revision.
- Stack: Python, TensorFlow/Keras, DWT, CNN.
- Handwritten Math OCR — a CRNN + CTC model (TensorFlow) that reads handwritten math expressions from photos and solves them, trained on a Kaggle dataset plus synthetic data.
- Portfolio Moon — personal portfolio website with a Matrix-style terminal landing page.
- Portfolio: mehulsri07.github.io/Portfolio-Moon


