CS graduate building real-time speech systems and shipping Flutter apps people can actually install.
I like work you can measure and ship. Streaming ASR → MT → TTS over WebRTC with hop-level latency. Flutter on the Play Store, not just in a repo. Backends that run in Docker Compose, not in a screenshot.
Currently: LoRA fine-tuning Whisper for Hindi–English code-mixed ASR, then serving it from a Rust ONNX runtime.
B.Tech — Computer Science and Business Systems, SSTC Bhilai · graduated April 2026.
speech-relay — real-time speech-to-speech translation
faster-whisper · IndicTrans2 / OPUS-MT · Coqui / edge-tts · aiortc · ONNX Runtime
Live EN→HI audio over WebRTC. 2505 ms mean / 2580 ms p50 end-to-end, instrumented per hop. Silero-VAD vs fixed-window flushing cut latency 23.6%. One shared ASR stream fans out to N listeners with independent translation targets (1.56× parallel efficiency at 2). ONNX fp16/int8 benches up to 11.3% faster, plus a WER/CER harness.
speech-relay-rust — Indic code-mixed ASR + Rust inference
Whisper · LoRA / PEFT · ONNX Runtime · Rust (ort) · MUCS
Two-part systems project: LoRA on Whisper for Hindi–English code-mixed speech (MUCS / OpenSLR-104), then an ONNX export path into a Rust inference runtime. In progress.
MediaPipe AI — distributed media pipeline
React · Express · Python · Redis · PostgreSQL · MinIO · Docker Compose
Five-container setup: Whisper transcription + BART summarization behind a queue, with an Express gateway, structured logging, and async job polling.
ByteWise — Flutter learning app · Play Store
Flutter · Riverpod · GoRouter · Drift / SQLite · Google Sign-In · Material 3
Shipped to Google Play. 788+ lessons across DSA, system design, and six languages. Offline-first two-tier cache (in-memory + Drift) so startup I/O stays cheap.
Nen — Flutter music player · Play Store (closed testing)
Flutter · just_audio · Riverpod · GLSL · CustomPainter
Full-featured player with a real-time GLSL audio visualiser on a Flutter canvas. Handled APK signing and a staged Play Console rollout.
Speech / ML — faster-whisper (CTranslate2), Silero-VAD, IndicTrans2 / OPUS-MT, Coqui / edge-tts, ONNX quantization, asyncio streaming, WER/CER harnesses
Mobile — Flutter, Riverpod, Provider, BLoC, Drift / SQLite, GLSL
Backend & data — FastAPI, Express, PostgreSQL, SQLite, Firestore, MongoDB, Redis
Infra — GCP (Cloud Run, Cloud Storage, Firebase), Docker Compose, GitHub Actions
