Keshav Raj

Software Engineer

Kochi, Kerala, India1 yr 11 mos experience
AI EnabledAI ML Practitioner

Key Highlights

  • Expert in building scalable AI orchestration platforms.
  • Reduced production container size by 75%.
  • Led successful data migration with zero issues.
Stackforce AI infers this person is a Backend-focused Software Engineer specializing in SaaS and AI technologies.

Contact

Skills

Core Skills

A2aGoLanggraphPostgresql

Other Skills

PythonRedisMicroservicesKubernetesMongoDBReact.jsAgentic AI DevelopmentLarge Language Models (LLM)Docker ProductsAmazon ECSNode.jsWeb DevelopmentMySQLDatabase Management System (DBMS)Object-Oriented Programming (OOP)

About

Hi! I'm Keshav , Software Engineer at IBM building production AI-agent orchestration platforms deployed across SaaS, IBM Cloud, and on-premises environments. https://keshavraj77.github.io/Portfolio/ I've implemented the full A2A (Agent-to-Agent) protocol stack in Python and Go — both as a client (multi-turn conversations, async push-notification callbacks via Redis Streams) and as a server (JSON-RPC endpoints, lazy task creation, dual-signal cancellation). I also built the protocol-translation microservice in Go that bridges our internal chat format to A2A JSON-RPC, shipped to SaaS and on-premises Kubernetes. Other highlights: • Implemented a LangGraph ReAct orchestration engine eliminating a separate planning phase for fast-inference LLM models • Built a progressive-disclosure skill system that reduced base system-prompt tokens by 82% and measurably cut LLM hallucination caused by context overload • Led a multi-data-center MongoDB to PostgreSQL migration of millions of records, including a custom character-sanitisation pipeline • Reduced production container image from 6 GB → 1.5 GB (75%) via multi-stage builds Outside IBM, I built AgentComm — a visual multi-agent orchestration desktop app with a drag-and-drop workflow builder (swarm pattern), MCP server integration, LangGraph-based parallel execution, and full A2A + multi-LLM support. College topper in IT at CUSAT (GPA 9.65) · IBM Hackathon 1st place. I enjoy solving backend challenges and building efficient systems that scale with ease.

Experience

1 yr 11 mos
Total Experience
1 yr 11 mos
Average Tenure
1 yr 11 mos
Current Experience

Ibm

2 roles

Software Developer

Jul 2024Present · 1 yr 11 mos · Hybrid

  • Built and scaled core components of an AI agent platform across SaaS, IBM Cloud, and on-prem. Implemented an end-to-end Agent-to-Agent (A2A) stack in Python and Go, including a LangGraph→JSON-RPC adapter, async callbacks via Redis Streams, and a hybrid task-state classifier. Developed server-side JSON-RPC endpoints with lazy task creation, deterministic artifact IDs, and sub-millisecond distributed cancellation.
  • Built a Go-based protocol translation service (Gin) bridging internal chat systems to A2A with SSE + sync support, Redis eventing, JWT auth, and OpenAPI specs.
  • Designed a LangGraph ReAct engine with a single-step reasoning loop, reducing latency and preventing cross-agent context bleed via isolated execution threads.
  • Created a progressive-disclosure skill system that reduced prompt tokens by 82% and per-turn usage by 56%, improving accuracy and reducing hallucinations.
  • Led migration of millions of records from MongoDB to PostgreSQL across data centers with zero issues, solving encoding challenges via a custom sanitization pipeline.
  • Also built a plugin framework (RBAC, PII masking, audit logging), reduced container size by 75%, implemented multi-tenant DB isolation, automated data retention, and a PostgreSQL-backed memory layer.
GoA2APythonLangGraphPostgreSQLRedis+2

Software developer Intern

Jan 2024Jul 2024 · 6 mos · Hybrid

PythonReact.js

Education

Cochin University of Science and Technology

Bachelor of Technology - BTech — Information Technology

Jan 2020Jan 2024

Stackforce found 100+ more professionals with A2a & Go

Explore similar profiles based on matching skills and experience