Final-year Computer Science student at Manipal Institute of Technology, Bengaluru, focused on backend engineering with Java and Spring Boot. I build production systems end to end — a live URL-shortener product, an OpenAI-compatible LLM gateway, and a real-time image-streaming application — with an emphasis on clean architecture, concurrency, and performance. Co-founder and founding engineer at GoLinkGone.
Co-founder and backend engineer. Short links, click analytics, and customizable QR code studio, serving real users. 27K+ pageviews in just ~1.5 Months · 500+ Sign-Ups · 550+ short links created · 22k+ redirects served. Ranked #9 of 100+ products on Peerlist (Week 26); also launched on Product Hunt.
- Customizable QR codes — branded QR generation driven by a validated JSONB style configuration, persisted in custom storage and re-rendered on demand.
- Analytics pipeline — a 10K-capacity click queue drained every 3 seconds via map/reduce into batched upserts, with murmur3-hashed visitor IDs on a virtual-thread executor.
- Platform — Caffeine caching, ES256 JWT auth, Bucket4j rate limiting, MaxMind GeoLite2 geolocation, host-based dual-domain routing, Supabase auth with Google SSO.
A single endpoint that routes, load-balances, and fails over across multiple LLM providers (Ollama, Gemini) while presenting the OpenAI wire protocol to clients, so standard SDKs and tools work unmodified.
- Reactive core — Spring WebFlux with R2DBC; SSE streaming with cross-provider chunk normalization.
- Routing & resilience — power-of-two-choices load balancing on live in-flight depth, TTFT-based provider health states, circuit breakers with connect-time failover.
- Security — SHA-256 gateway keys, AES-GCM-256 provider credentials, connect-time SSRF guard, per-key rate limits and spend caps.
- Verified — ~0.3 ms p50 gateway overhead, zero errors at 200 concurrent, 93 tests including failure injection.
Rebuilds the subject of one image using the pixels of another, streamed to a WebGL client and played back as an animated particle system.
- AI subject extraction — U²-Net saliency model served via ONNX Runtime.
- Streaming — custom binary WebSocket protocol (32-byte header + 256 KB chunks) feeding a Three.js instanced-points renderer with custom shaders.
- Concurrency — dual-lane processing pipeline with buffer pooling, a global concurrency semaphore, and per-IP rate limiting.
- Tried and loved by 100+ people.
Languages
Backend
Data
Infrastructure & Tools
Codeforces Expert · LeetCode Guardian (2142) · active on CodeChef and AtCoder.
- 1st place — MAHE CodeWars 2026 (3rd place, 2025)
- Top 5 Finalist — Hitachi Visisonics AI'26
- Preparing for ICPC regionals
