Ayush Kumar

I build queue-driven backend systems in Python and FastAPI, including pyFlow, a distributed workflow platform that processes CSV workloads asynchronously through Celery and Redis.

With pyFlow, I designed a DAG engine with retries, cycle detection, dependency-aware execution, and 14 REST endpoints backed by SQLAlchemy and Alembic. The platform achieved roughly 70x faster submission than inline processing for a 25 MB CSV workload and sustained approximately 885 jobs per minute on four workers.

I also built docXpo, a multi-LLM backend that unifies Anthropic, Gemini, OpenAI, and Ollama behind a single provider interface. I implemented streaming SSE responses, RAG with pgvector and HNSW, Redis semantic caching, rate limiting, and cost and latency metrics.

As an open-source contributor, I've had 28+ pull requests merged across nine production repositories at AiiDA, the Apache Software Foundation, OSGeo, and pvlib. I enjoy working through maintainer review, CI pipelines, testing, and production-quality fixes.