Vibhor Bhatnagar.
// systems ยท distributed networks ยท AI workflows

I design databases that don't lose writes, APIs that don't block threads, and LLM pipelines that don't hallucinate.

I am a Software Development Engineer with 5.5+ years of experience engineering transaction-heavy backends and automated AI workflows. Currently at Amazon, I scope and deliver scalable merchant-facing acquiring systems, build serverless messaging infrastructures, and develop production-grade agentic data pipelines.

"Fuzzy middle, strict edges. Let the model handle the single ambiguous step it is good at, wrapped in plain, boring, deterministic code you can verify."

Recent Work & Blueprints

Co-designed and launched merchant acquiring pipelines in Japan, managing integration with external banking networks. Developed serverless transaction systems capable of handling peaks of 18,000 requests/minute (300 TPS) under strict sub-second SLA bounds.

Java / AWS Lambda / Fargate / SQS / DynamoDB

An isolated Python/FastAPI semantic search pipeline implementing an Agentic ETL ingestion sandwich (strict regex parsing + LLM chunking + validation schemas) and a bounded Verify-Loop to evaluate and correct hallucinated citations against raw database source segments.

Python / FastAPI / Elasticsearch / Pytest / Ollama

A Maven-based multi-threaded document crawler and parser employing graph cycle checks, Redis Bloom filters for O(1) duplicate checks, and a Kafka messaging stream to route high-frequency crawl events safely into Postgres.

Java / Maven / Spring Boot / Apache Kafka / PostgreSQL