First-year Information Technology student exploring full-stack engineering, machine learning pipelines, and real-time systems. Not an expert, just curious, consistent, and building things that solve real problems.

I document my journey through projects here on GitHub and share the behind-the-scenes on Instagram.

Languages

Frontend

Backend

Databases & Infrastructure

ML & Data

DevOps & Deployment

https://github.com/mishradwaterlaw/echosort

Event photo sharing with facial recognition. Upload a selfie, get back every photo of you from the entire event gallery.

A full-stack application built from scratch across five deployment platforms. Attendees visit a shared event link, take a selfie, and the system uses ArcFace face embeddings and pgvector cosine similarity search to find every matching photo in the event gallery.

What makes it non-trivial

• Decoupled ML pipeline: photo uploads are queued via Redis, processed asynchronously by a TensorFlow worker, and stored as 512-dimensional vectors in a pgvector HNSW index
• Secure by design: private storage buckets, time-limited signed URLs, Row Level Security on every table, and JWT verification using getUser() not getSession()
• Server Actions pattern: credentials never touch the browser bundle
• Real per-file upload progress via XHR (Fetch API does not expose upload progress)

Stack: Next.js 14 · FastAPI · DeepFace/ArcFace · Supabase (PostgreSQL + pgvector + Auth + Storage) · Upstash Redis · Modal.com · Vercel · Render

• Vector databases and semantic search: pgvector, embedding models, approximate nearest neighbor indexing
• ML deployment: Serverless GPU workers, model serving, pipeline optimization
• DSA in C++: Strengthening problem-solving foundations for technical interviews
• Open Source contribution: Learning by reading and improving production codebases