कल्याण

Hello

कल्याण · Bangalore, IN

Kalyan

Software × AI. Researcher. Builder.

Software Engineer I at Guidewire · where AI meets software

research papers before graduation, production AI after — I build systems that think, remember, and ship. everything else is configuration.

Selected work ↓ResearchSay hello
Kalyan

Experience

Research first, production after

  1. Feb 2025 – present

    Software Engineer I · Guidewire

    • Reduced LLM claim-summarization latency through end-to-end payload optimization.
    • Developed a multi-judge LLM evaluation framework (5 models, 5 metrics per field, 95% confidence intervals).
    • Shipped a production claim-summarization agent with data model, persistence and REST API.
    • Built a production event-scheduling agent (LangGraph workflow with Slack, ServiceNow, room booking).
    • Delivered a real-time voice claims-intake prototype and a voice MCP capability.

  2. Sep 2024 – Jan 2025

    Research Intern · IISc Bangalore

    • Built deep learning models for underwater object detection using VLMs on sonar imagery.
    • Co-authored the TRAM paper on transformer-based Mask R-CNN for sonar data.

  3. May 2024 – Aug 2024

    AI/ML Intern · Hexon Global

    • Reduced LLM API costs by about 20% through model-call optimization.

  4. Jul 2023 – Jul 2024

    Student Researcher · Amrita University

    • Developed MedGPT diagnostic AI (accuracy improved from 30% to 80%).
    • Co-authored 4 research papers across medical AI, computer vision and NLP.

Selected work

Systems that think, remember and ship

agam

Featured

Open-source memory system for AI coding agents — proactively injects relevant context instead of retrieving it on demand. 300+ tests, sole author and maintainer.

PythonMCPSQLite FTS5
GitHub →

Voice FNOL

Featured

Voice-based insurance claim intake: a parallel LLM pipeline (conversation + extraction), local STT, on-device TTS and a 3D audio-reactive visualizer.

TypeScriptFastifyClaude APIThree.js

Cognitive OS

Multi-agent Claude Code setup with 30+ custom skills and 14 specialized agents, a SQLite knowledge graph of 400+ entities, session hooks and persistent memory.

Claude CodePythonSQLite FTS5MCP

MedGPT

Diagnostic AI deployed at AIMS Kochi Hospital; accuracy improved from 30% to 80%.

FlutterOpenAIAWSRedis

Kanaku

AI-powered dead-stock management for FMCG distributors — 221 real products, 74 stores, statistical ML scoring and WhatsApp integration.

Next.jsPrismaGeminipgvector
GitHub →

SheLaw

Legal assistance platform for women using RAG with multi-query retrieval.

FlutterLangChainLangGraph
GitHub →

JobScrapper

AI job matching with two-layer scoring (heuristic + semantic), automated scraping and email alerts.

PythonNext.jsOpenAINeon Postgres

VoiceAI

Automated customer interactions — product explanations and event registration — integrated with Indian IVR systems.

PythonTTSWebSockets

Core competencies

What I do

01

Agent architectures

Multi-agent orchestration, tool integration, memory systems and voice pipelines with LangGraph, MCP and Claude Code agents.

02

AI evaluation & quality

Evaluation frameworks, prompt versioning, LLM-as-judge, CI/CD for agents with confidence intervals and rubrics.

03

Enterprise AI integration

MCP servers, knowledge search, insurance claim processing, ClaimCenter and InsuranceNow integrations.

04

Cognitive tooling

A knowledge graph of 400+ entities, 30+ custom skills and persistent memory systems.

05

Research to production

Five published papers deployed across hospitals and enterprise users.

06

Workshops & enablement

Agent-building workshops, talks on AI agents & MCPs, ML Bootcamp co-leadership (500+ participants).

Milestones

1st placeTFUG x Kaggle Hackathon2023
2nd placeGender x GenAI Hackathon2025
6th placeIdea Ignite Hackathon / Codebyte2024
Organizer & panelistDevSummit2025
Co-led ML Bootcamp500+ participants2024
CertifiedGuidewire Certified Associate, Jutro Developer2025
Technical stack →