I architect high-throughput asynchronous microservices, distributed messaging topologies, and production machine learning pipelines. My engineering focus centers on eliminating synchronous bottlenecks with non-blocking I/O (asyncpg, aioboto3), replacing polling loops with broker-based push streams (RabbitMQ, MQTT), and deploying low-latency vision, speech, and retrieval models across Google Cloud, AWS, and Cloudflare.
- Led offshore engineering spanning Google Cloud and AWS for US social video media brands (The News Movement, Recount).
- Architected keyless cross-cloud authentication using AWS-to-GCP Workload Identity Federation, allowing AWS Lambda services to query BigQuery catalogs via REST APIs without static service account keys.
- Decomposed monolithic pipelines into isolated Cloud Functions with LLM classification and guarded SQL MERGE write-backs to preserve editorial annotations.
- Engineered an automated video transcription pool featuring time-budgeted execution, exponential jitter backoff, and idempotent BigQuery job management.
- Replaced polling and SSE with a RabbitMQ topic exchange and Web-MQTT gateway over WebSockets, delivering real-time progress for 5 event domains with role-based topic filtering.
- Integrated GPU-accelerated microservices: polygon instance segmentation via SAM 2.1 and multi-object tracking via YOLO11x with BoT-SORT.
- Built browser-direct presigned S3/MinIO multipart uploads with SHA-256 duplicate detection and resume checkpoints.
- Engineered a 9-stage Celery ingestion pipeline incorporating Surya OCR, chunking algorithms, and hybrid dense/sparse vector indexing in Qdrant.
- Led a non-blocking asynchronous migration across 3 core microservices using SQLAlchemy 2.0 (asyncpg), aioboto3, and httpx to eliminate event loop starvation.
- Integrated self-hosted Faster-Whisper (STT) and Kokoro-82M (TTS) microservices with TTL caching for low-latency speech pipelines.
- Built an end-to-end processing pipeline cutting scanned Hindi newspaper pages into individual articles using polygon refinement and Surya OCR on vLLM.
- Implemented OCR hallucination defence using zlib compression ratio checks (<0.06) and temperature escalation.
- Configured an OpenSearch index featuring ICU transliteration and custom Hindi analyzers.
| Repository | Focus Area | Technology Stack |
|---|---|---|
| kokoro-tts | Self-hosted, containerized Text-to-Speech API with interactive playground | Python, FastAPI, Kokoro TTS, Docker |
| speech-to-text | Speech recognition service with WebSocket streaming and batch endpoints | Python, Faster-Whisper, WebSockets |
| flare | Edge-native real-time conversational voice companion | TypeScript, Cloudflare Workers, WebRTC |
| portfolio | Statically generated developer portfolio with strict CSP and zero runtime overhead | Astro, TypeScript, Vanilla CSS, Cloudflare Pages |
| Domain | Technologies & Frameworks |
|---|---|
| Languages | Python, TypeScript, JavaScript, SQL |
| Applied AI & ML | vLLM, Surya OCR, SAM 2.1, YOLO11x, BoT-SORT, Faster-Whisper, Kokoro TTS, Qdrant, LangGraph, RAG |
| Backend & APIs | FastAPI, SQLAlchemy (Asyncpg), Celery, Pydantic v2, Hono, Uvicorn, REST, JWT, OAuth2 |
| Messaging & Streaming | RabbitMQ (Web-STOMP), MQTT (EMQX / Web-MQTT), WebSockets, Server-Sent Events, aio-pika |
| Data & Storage | PostgreSQL, Redis, BigQuery, OpenSearch, MinIO, AWS S3 (aioboto3), DynamoDB, Cloudflare D1 |
| Cloud & DevOps | Google Cloud (Functions, Workflows), AWS (CDK, Lambda, S3, Cognito), Cloudflare Pages/Workers, Docker, Linux, Terraform |
| Testing & CI/CD | Pytest, Vitest, Playwright, GitHub Actions, ESLint, Prettier |
- Bachelor of Computer Applications (BCA)
SSG Pareek College, University of Rajasthan (2022 – 2025) · Jaipur, India