I care about systems that are correct before they're clever.
I build things end-to-end — from signaling servers and NLP pipelines to 3D visualizers. Most of my time goes into understanding why a system breaks, not just how to patch it. I prefer fewer, deeper projects over a portfolio of tutorial clones.
Right now I'm going deeper into distributed system design and exploring how ML models can be embedded into real products rather than living in notebooks.
|
Secure Real-Time Remote Desktop Platform Cross-platform Electron app with peer-to-peer screen sharing, remote control, and chunked file transfer across 4 dedicated WebRTC data channels. ECDH P-256 key exchange + AES-256-GCM E2E encryption. 38-endpoint REST API, JWT + Google OAuth2 + TOTP 2FA, deployed on AWS EC2/RDS with a self-hosted TURN server.
|
Full-Stack Legal Research Intelligence Platform 5-stage NLP pipeline: domain classification → InLegalBERT/BioBERT embeddings → FAISS nearest-neighbor search → coherence detection → severity-ranked conflict reports. Flask + pgvector backend, 15+ REST endpoints, NetworkX citation-graph with NLI relationship classification, Next.js frontend with RAG chat.
|
|
3D Semantic File Exploration Platform Multi-stage NLP pipeline: TF-IDF extraction → SentenceTransformer embeddings (all-MiniLM-L6-v2) → custom PyTorch genre classifier — auto-classifies 1,000+ documents by semantic similarity. Results surface in an interactive 3D Canvas visualizer. Electron + Python NLP backend + PostgreSQL supporting 10,000+ records with full-text search.
|
Real-Time Sorting Algorithm Visualizer C++17 + SFML desktop app that renders 6 sorting algorithms (Bubble, Selection, Insertion, Merge, Quick, Heap) bar-by-bar with live comparison/swap counters. Colour-coded states: amber = comparing, red = swapping, purple = pivot, green = sorted. OCP-compliant design — new algorithms plug in via a single header, zero changes elsewhere.
|
Languages
Backend & Infra
Frontend
ML / NLP
