Madhusudan Singh — Backend & Distributed Systems Engineer

95% codebase reduction 12,000 → 600 lines

9× faster API response REST → gRPC, org-wide

1M+ rows per file parallel bulk-I/O layer

0 lost rows under load degrades to delayed

Skills

Languages: JavaPythonRustC/C++TypeScriptSQL
Backend & systems: Spring BootgRPCTemporalDistributed systems designEvent-driven architectureResilience4jExactly-once semanticsConcurrency
Data & cloud: MySQLDynamoDBS3AWS LambdaEC2DockerKubernetesCI/CD
AI & agents: LangGraphLangChainRAGDSPyPyTorchPinecone/Milvus

Mar 2023 — Present

JAVIS Technologies·Bengaluru ·Started as an intern

Led the ingestion compiler — per-tenant config compiles into an immutable execution profile, replacing 12,000 lines of per-client code with 600.
Built the durable execution engine on Temporal — crash-resumable mid-file, with version-pinned profiles for reproducible replay.
Shipped the async framework now embedded in every backend service across the org: idempotency, leases, and crash recovery, so domain teams write only handlers.
Moved REST to gRPC for a 9× response-time improvement, and replaced the ORM with a bulk-I/O layer to scale past 1M rows per file.

Deep dives

JavaSpring BootTemporalgRPCRustMySQLAWS Lambda

Nov 2024 — Dec 2025

Endura AI / Aureus AI·Bengaluru

Architected a LangGraph multi-agent system for chartered accountants, automating bookkeeping and month-end close with direct Tally integration.
Built a tenant-isolated RAG pipeline with DSPy/Pydantic-enforced extraction, and a reconciliation agent that cut month-end close from 7–8 days to 3–4 hours.

Deep dives

PythonLangGraphDSPyGeminiRAG

Things I built on my own time.

A stacked GAN generating photorealistic birds from text descriptions

A real-time prediction market where humans compete against AI agents

Education

B.Tech, Electrical Engineering
IIT Roorkee · 2019–2023

Competitive programming

The fastest way to reach me is email. I read everything, and I reply to anything that isn't a template.