Prateek Aryan
Senior Backend Engineer building distributed systems and platform automation on Python, FastAPI, and AWS. Based in Bengaluru · Comfortable across US and EU overlap.
About
I build backend systems that stay up when things get hard.
Currently Senior Software Engineer II at UsefulBI, where I architected the backend for an AI-powered clinical document platform on FastAPI + PostgreSQL — designed for FDA 21 CFR Part 11 compliance from day one, with audit trails, section-level versioning, and RBAC across 12+ modules.
Most recent measurable win: I cut a clinical document retrieval API from 7–8 seconds to under 50ms — a 99.3% reduction — by replacing synchronous S3 polling with a database registry and turning O(n) prefix scans into O(1) indexed queries. Full write-up here.
Before UsefulBI, I spent five years at Cisco Systems building the platform automation that kept 200+ enterprise applications reliable: MySQL upgrades, Postgres provisioning, Oracle switchover, production-to-non-prod refresh workflows, MongoDB cluster operations, and a resilient switchover system using gRPC across primary/standby clusters. Along the way I cut SOX compliance reporting from 3+ hours to under 10 minutes and reduced manual DBA workload by 80%.
Recently I've been going deeper on LLM orchestration — designing automated report generation pipelines with AWS Step Functions and Bedrock, with Lambda handling each stage (ingestion, section-wise generation, validation, assembly) and domain SMEs collaborating on regulatory requirements.
Certified: MongoDB DBA · AWS Certified Cloud Practitioner · HashiCorp Terraform Associate · PCAP — Python Programming.
Education: B.E. Computer Science, BMS College of Engineering (9.2 GPA).
Case studies
Cutting a clinical document retrieval API from 7s to 50ms
How I turned an O(n) S3 prefix scan into an O(1) indexed PostgreSQL query — with a four-phase migration plan that caught a latent bug before it became an incident.
Orchestrating multi-step LLM workflows with AWS Step Functions and Bedrock
Parallel section-wise generation, validation with repair prompts, human-in-the-loop via Step Functions callback tokens, and the observability layer that makes it all debuggable at 2 a.m.
Building resilient switchover automation with gRPC
How I replaced a brittle REST-based primary-standby failover system with a gRPC-driven state machine that handled hundreds of enterprise clusters without split-brain — and why fencing is non-negotiable.
Accelerating SOX compliance reporting from 3+ hours to 10 minutes
How I replaced a quarterly spreadsheet-wrangling exercise with an automated pipeline that auditors actually liked. Snapshot first, process second, evidence table in the middle.
Automating MySQL cluster upgrades from 1 hour to 10 minutes as an intern
My first real production automation project — shipped on the side in my fourth month as an intern, before anyone asked for it. The lessons I learned shadowing senior engineers, classifying step outcomes, and the five patterns I still use four years later.
Open-source projects
- documind Production-grade RAG Q&A API — FastAPI + pgvector + Celery + SSE
- deep-focus-studio Neuroscience-backed ambient sound & focus timer (live)
- dj-visualizer Real-time mic-reactive 3D WebGL visualizer
- mcp-servers Collection of MCP servers with shared transport and OAuth
- github-mcp MCP server wrapping GitHub CLI for Claude integration
Contact
prateekaryyan@gmail.com · github.com/prateekaryann · linkedin.com/in/prateek-aryan
Focus areas: distributed systems, platform reliability, LLM infrastructure, GenAI backends.