Ishan Dilmith

Architecture

Modular FastAPI backend split into four domain modules (live coaching, role-play, adaptive training plans, analytics) that share typed contracts. A React + Vite SPA runs MediaPipe in the browser and streams signals over WebSockets; data lives in Supabase Postgres with Alembic migrations, shipped as Docker images through GitHub Actions behind Nginx.

Architecture

Fat-service design: every business rule (booking windows, battery-slot capacity, reservation lifecycle) lives in an ASP.NET Core Web API over MongoDB with JWT auth. The React web app and the Kotlin + Jetpack Compose Android app are thin REST clients, so both always behave identically.

Architecture

Next.js client, Express API and a separate FastAPI ML risk-scoring service. My work sat in the auth layer and security pipeline: rate limiting, OTP and login hardening, Google OAuth with PKCE, plus SAST audits and OWASP ZAP scans with the findings fixed.

Architecture

Node.js / Express microservices for auth, restaurants, menus, orders, delivery, payments and notifications, all reached through a single API gateway. Socket.io pushes live delivery updates; the stack runs on Docker Compose locally, with Kubernetes manifests and an Nginx-served React frontend for production.

Architecture

A CrewAI pipeline of four agents (file reader, rubric evaluator, feedback writer and grade reporter), each with its own task and tool, running in sequence on a local Ollama model (Qwen 2.5 7B). Reads PDF and DOCX submissions, logs every run, and is covered by pytest unit tests.

Architecture

Two-stage notebook pipeline: preprocessing removes duplicates and scales the Time and Amount features into a cleaned dataset, then four classifiers are trained and tuned on it, saved as .pkl artifacts, and their metrics exported to CSV for side-by-side comparison.