Skip to content

Hi, I'm

Indira Kumar

I build the infrastructure agents run on. 10,000+ people use it every day.

AI Engineer 2 @ Thermo Fisher Scientific

here are some things I believe in

Experience

Thermo Fisher Scientific

AI Engineer 2 · May 2024 – Present · Bangalore

  • Built a sales assistant that turns natural-language requirements into full instrument and parts quotes by two paths: a LangGraph multi-agent workflow and a custom deep-research agent; $28M projected revenue, not yet realized.
  • Took a RAG service assistant from concept to production in 2 months for 3,000+ field engineers, saving $3.5M a year.
  • Defined the org's MCP server archetype and safety requirements, then shipped two of its earliest production servers.
  • Raised indexing throughput 7x (2 repos/week to 2/day) by automating ingestion of 400 repositories from SharePoint, S3, Bynder, and Fluidtopics.
  • Cut p95 response time 53% (26.5s to 12.4s) and closed a P1 incident by parallelizing document and ticket retrieval.
  • Built a customer intelligence assistant surfacing upsell opportunities across existing and prospective accounts; $3.1M incremental annual revenue.
LangGraphLangfuseMCPFastAPIDatabricksAzure OpenAIAzure AI SearchPostgreSQLDynamoDBAWSAngular

Pancakes

Founder · May 2025 – Dec 2025 · Remote

  • Shipped Reflect, a feedback widget with a 3-line integration that syncs into JIRA, Linear, and GitHub as tickets. MVP to 20+ DAU in 3 weeks, ran 7 months in production, then sunset on retention data.
  • Shipped Saplings, a React Native education app, piloted it in a school, and pivoted on market data. Two products end to end in one founding stint.
  • FastAPI, React, Terraform on GCP and AWS RDS. 40% lower API latency via Celery and RabbitMQ; sub-500ms global load via Cloudflare CDN.
FastAPIReactReact NativeTerraformGCPCloudflareCeleryRabbitMQ

Dreamimbue

Technical Co-founder · Apr 2023 – Aug 2023 · Remote

  • Built and scaled a finance-management PWA for small financial institutions to 300+ DAU, leading a team of 3. Still running 2 years later on recurring revenue.
  • Two microservices on Node.js and Spring Boot with PostgreSQL and MongoDB, plus multi-role JWT auth for five user types.
Node.jsSpring BootPostgreSQLMongoDBReact

Projects

Agent infrastructure

mem-alive

A memory layer for AI agents, on PyPI. Semantic, episodic, and procedural stores behind one federated recall(), over pluggable storage and embedding backends. Fully async, tested per module. v0.1.0, MIT.

PythonasyncioembeddingsPyPI

Jarvis

A personal assistant you delegate tasks to, built around a configurable human-in-the-loop permission gate per task. 17-scenario eval suite across email, calendar, browser, memory, and research; a local Qwen3 14B passes 13 on the last full run.

agent harnesspermissionsevalsQwen3 14BOpenAI API

codesmith

A CLI coding agent harness in TypeScript. Built to understand the tools I ship with: roughly 90–95% of my output is agent-driven, from spec to plan to TDD to PR review.

TypeScriptCLIagent harness

Products

Daily Cricket

Daily Cricket

A daily NYT-style cricket guessing game. Six puzzles refresh every morning at 5:30 AM IST. 4,000+ MAU, all organic. Custom scraping pipelines normalize cricket stats to generate fresh puzzles daily.

Next.jsSupabaseTailwind CSSPostHog
LocalPDF

LocalPDF

A browser-based PDF toolkit. Merge, split, compress, rotate, and sign PDFs, fully client-side via Web Workers. Nothing leaves the browser.

Next.jsTypeScriptWeb Workerspdf-libPDF.js

Skills

AI

LangGraph
multi-agent orchestration
RAG
MCP
agent memory
LLM evals
Langfuse
OpenAI API
Azure OpenAI
Databricks Vector Search
Azure AI Search

Workflow

Claude Code
Codex
spec-first
agent-driven TDD
test and eval review pre-PR
codesmith

Languages & Backend

Python
TypeScript
SQL
Java
FastAPI
Node.js
Spring Boot
Celery
RabbitMQ

Frontend & Infra

React
Next.js
Angular
Tailwind
PostgreSQL
DynamoDB
Redis
AWS
GCP
Azure
Docker
Terraform
CI/CD