Hi, I'm
Utkarsh Sharma
Building distributed systems, data pipelines, and AI-native infrastructure.
➜ ~ whoami
Utkarsh Sharma
➜ ~ role --current
Software Engineer
➜ ~ location
Bangalore, India
➜ ~ ▍
About
Software Engineer with production experience across payments infrastructure, big-data pipelines, and applied AI. I've re-architected batch systems processing millions of transactions, replaced legacy ingestion pipelines with modern connectors, and designed multi-agent AI systems using the Model Context Protocol. I care about systems that are fast, observable, and hard to break.
Indian Institute of Technology (ISM) Dhanbad
B.Tech in Electronics and Communication Engineering · Dec 2020 — May 2024
Experience
Software Engineer · Visa
Aug 2025 — May 2026Bangalore, India
- Led the migration of a Java/Spring Boot batch processing service from V1 to V2 APIs, re-architecting the core application to support the new contract across 25+ integration points while processing 2M+ transaction records per batch cycle.
- Refactored the core application codebase, cutting duplicated logic by 40% and reducing end-to-end batch runtime by 30%.
- Validated the migrated service across dev environments and automated deployment through a Jenkins CI/CD pipeline, reducing release time from 45 minutes to 12 minutes.
- Engineered a Sqoop-to-Apache SeaTunnel migration for a Hive-based Hadoop pipeline, building a custom SeaTunnel source, SQL connector, and sink that replaced the legacy ingestion path for 35+ jobs and improved ingestion throughput by 45%.
Software Engineer — Language AI Framework & MDE · Samsung Research and Development Institute
Jul 2024 — Jul 2025Bangalore, India
- Key contributor to a multi-agent POC for Bixby enhancement in Python, integrating external capabilities over the Model Context Protocol (MCP) via a WhatsApp MCP server.
- Managed and optimized Bixby capsules using JavaScript; added training data that improved command recognition accuracy by 35%.
- Designed Bixby agents using established software design patterns (Factory, Observer), reducing code redundancy by 40%.
Projects
A governed gateway for the Model Context Protocol that aggregates multiple upstream MCP servers behind a single OAuth 2.1-authenticated endpoint.
- Implemented Streamable HTTP transport with session resumption and protocol version negotiation.
- Capability-level authorization with OPA/Rego policies and a KMS-backed credential broker — upstream credentials are never exposed to AI clients, mitigating confused-deputy and token-passthrough attacks.
- Rug-pull detection via cryptographic pinning of tool definitions, plus bidirectional scanning that redacts PII and instruction-injection patterns from tool outputs.
- Provisioned the full AWS footprint (EKS, RDS, KMS, IRSA, Route53) as modular Terraform, with a Java Kubernetes operator exposing MCPServer CRDs for GitOps-managed server registration.
A distributed pub-sub messaging system built for high availability and consistent state management under load.
- Apache ZooKeeper for automated leader election and topic synchronization.
- Aerospike for rate limiting (500+ requests/sec) and RabbitMQ with the Akka actor model for concurrent message processing — 40% throughput improvement while maintaining fault tolerance.
- Lightweight real-time messaging via MQTT, reducing latency to 5ms for IoT devices across heterogeneous systems.
Skills
Languages
Backend & Frameworks
Data & Messaging
AI / ML
DevOps & Tools
Cloud & Infra
Core CS
Contact
I'm open to interesting conversations about distributed systems, AI infrastructure, and everything in between. Reach out through any of these.