Curious developer. Community builder. Competitive coder.
I build full-stack systems across Go, Python, and TypeScript, with experience at Salesforce, the University of Queensland Centre of Excellence, and DRDO.
I contribute to Vajra, an open-source Python/Suricata intrusion-prevention system, focusing on reliable installation, detection rules, and IPv4/IPv6 traffic blocking.
B.E. in Computer Science and Business Systems, Thapar Institute of Engineering & Technology, Patiala.
Aug 2023 - July 2027 · CGPA 8.67/10
Oct 2026 – Present · Remote
May 2026 – Jul 2026 · Bangalore
Network Security & Infrastructure
Jan 2026 – Mar 2026 · TIET, Patiala
Jun 2025 – Jul 2025 · New Delhi
Shazam-like audio fingerprinting built from scratch in Go. A DSP pipeline combines STFT with Hamming windowing, spectral peak extraction, and 32-bit combinatorial hashing to identify songs from noisy 5–15-second clips. WebAssembly sends fingerprint hashes instead of audio (~98% less bandwidth), with a single batched indexed lookup and CLI, REST, and WASM interfaces.
Fully self-hosted RAG chatbot with zero external LLM API calls; inference runs locally via llama.cpp. In project measurements, a semantic response cache reduced repeat-query latency from 2.5s to 0.3s. History-aware query rewriting supports follow-ups, while an SSE gateway and Redis-backed async workers stream tokens with cancel-on-disconnect.
Go command-line tool that turns natural-language instructions into shell commands, with configurable AI providers and local Ollama support.
A working reimplementation of AWS Lambda's core runtime behaviour on local Docker containers. Hard-kill invocation timeouts (the host never trusts user code to exit), warm-container idle eviction via a background reaper, per-container event queues that stop cross-container event stealing, per-function warm pools so concurrent invokes fan out and callers block when the pool is saturated, and the Lambda Runtime API long-poll loop (/next then handler then /response). A FastAPI host process orchestrates the whole container lifecycle.
A persistent, agentic wellbeing and habit-planning assistant — deliberately not a chatbot with a database. It recalls your history and long-term memory, compares adherence with deterministic statistics rather than a model guess, explains what it noticed with concrete evidence, then proposes a plan change and waits for your approval. On approval it acts for real: a versioned plan update, an audit event, and a scheduled follow-up run by a background job that later adapts the plan or decides no intervention is needed. Next.js with Genkit/Gemini and Firebase auth. Non-clinical.
Kompte is a sports event hosting and management platform that allows users to create, manage, and participate in various sports events. It features user authentication, event creation, registration, and a leaderboard system as well as tournament bracket generation.
An automated video-dubbing pipeline with voice cloning and multi-language support. It transcribes with OpenAI Whisper, translates subtitles offline with Argos Translate, synthesises voice-cloned speech in the target language with Coqui XTTS v2, and reassembles a dubbed video with pitch-preserving time-stretch. Optional Demucs source separation keeps the original score under the cloned speech. Whisper and XTTS load once per run, artifacts are cached by audio fingerprint so failed runs resume, and the device (CUDA / MPS / CPU) is auto-detected. Ships as a one-command Docker image with a Gradio web UI.
Talkeys is a social-first event discovery platform that enables users to find, host, and attend events, while engaging in anonymous, chat-based communities. Integrated payment gateway and handles frontend and backend, integration and features.
ThaparGo is a ride-sharing platform designed specifically for Thapar University students, facilitating easy and secure carpooling. It allows users to post and find rides, ensuring a convenient travel experience within the campus community.
Languages, frameworks and infra I reach for.