AI Memory Layer
Hearly Knowledge
Developer-focused RAG infrastructure — a scalable memory layer for AI systems. Ingest, chunk, embed, retrieve and rerank, backed by pgvector with sub-200ms search.
Software Engineer · Bengaluru
Madan Hegde
I design and build scalable web applications and AI-powered tools, from idea to production.
Currently building workforce systems at PagarBook and AI knowledge infrastructure at Hearly. Also shipping a commissioned AI agent for a sitar maestro.
Open to interesting work
Selected work
AI Meeting Copilot
Hearly
Turns meeting audio into structured summaries, extracted tasks and searchable insights, with interactive Q&A over transcripts.
Personal AI Agent · In progress
Sitar Maestro
A solo, end-to-end agent — voice + text, contextual long-term memory, and tool calling for real-world scheduling and content tasks.
AI AgentsTool CallingVoiceComing soon
Experience · ~7 years
Nearly seven years building production web systems, developer platforms and AI-powered tools.
2025 —
PRESENT
PRESENT
Founder & Engineer Hearly FOUNDER
Building RAG-powered knowledge infrastructure for AI systems — with paying customers and an npm-distributed widget running in production.
Sub-200ms retrieval · 10-min integration · Shipped on npm
Knowledge Platform
Built the RAG pipeline end to end — ingest, chunk, embed, store, retrieve, rerank — on pgvector over Postgres.
Shipped an embeddable widget, published as @hearly-knowledgebase/widget on npm; developers integrate in under ten minutes.
Connected the knowledge base to LLM chat completions through structured retrieval, now powering paying customers in production.
Meeting Copilot · earlier product
Audio processing and structured extraction pipelines turning meeting recordings into summaries, tasks and searchable transcripts, with real-time intelligence over a WebSocket-driven UI.
Architecture notes ↓
Chose pgvector over Pinecone to drop a managed-service dependency and keep retrieval under 200ms — one Postgres to operate, one place for the data, and reranking close to storage rather than across a network hop.
AI · TypeScript · React · Node.js · npm
2022 —
PRESENT
PRESENT
SDE 3 Pagarbook
Leading Pagarbook's first AI integration — a tool-calling pipeline over the production backend that lets users take real actions on workforce data in natural language.
Expanded scope from frontend lead to fullstack ownership — now shipping across a Node.js/NestJS backend in addition to the web stack.
Designed the tool-calling layer so natural-language requests map onto existing production endpoints rather than a parallel API surface.
React · Next.js · Tailwind · Radix · Node.js · NestJS · Docker
2022Frontend Engineer @ Scenes
2020 — 2022Co-Founder & CTO @ Digiyogi Technologies
2019 — 2020Frontend Engineer @ Learnyst
Writing
Lesser-Known JavaScript Features — Part 1Medium ↗Lesser-Known JavaScript Features — Part 2Medium ↗Fast, maintainable, product-focused, simple — complex systems should feel simple to use.