# Malai — Muthhukumar Malaiiyyappan
Updated: October 2026
Website: https://malai-site.vercel.app

## Background
AI product engineer and former founder who turns ambiguous customer problems into production systems. Builds LLM, voice, retrieval, data, and full-stack products end to end, from customer discovery and architecture through deployment and iteration.

## Experience
- kGrid.ai / YaKnow — Co-founder & Engineer, Palo Alto, March 2024–September 2026. Built an AI interviewer used across 1,000+ expert calls, with dynamic follow-up questions and structured, searchable outputs. Built onboarding and automated identity verification for 2,900+ participants, an LLM-driven incentive engine, and Tango payouts totaling $6K+. Built transcript cleaning, claim extraction, and indexing for retrieval and downstream AI workflows.
- Unlock Labs — Co-founder & Engineer, Palo Alto, July 2023–March 2024. Built DialMe.at, a voice AI agent for SMB lead qualification. Used 3,000+ trial calls to refine conversations, qualification logic, and agent behavior.
- NFTHailer — Data Engineer, Bay Area, May–August 2022. Built and deployed automated ETL pipelines for recurring public-blockchain ingestion, reusable datasets, and on/off-chain product metrics.
- UrbanZoom — Data Analyst, Singapore, August 2019–December 2020. Analyzed approximately 468,000 housing transactions spanning 25 years in Python; property-return research was featured on Bloomberg. Managed over $20K/month in acquisition spend, ran A/B tests, worked with 11 realtors, reduced average cost per lead by approximately 30%, and automated lead handoff.

## Projects and research
- SonicAlpha: Podcast intelligence for public-market research. Built ticker-indexed discovery, speaker identification, LLM classification, embeddings/vector retrieval, and emerging-signal detection. Work includes query-rewrite and ticker-resolution evaluations, comparative model benchmarking, error analysis, and interactive evaluation dashboards. Also worked on podcast/YouTube ingestion and company/executive intelligence infrastructure. Alternative-data work was exploratory.
- RiffOn: Personalized, topic-specific insights from long podcast conversations, with daily digests. Reached 337K+ website users and delivered 100K+ newsletter emails. A distinct product from SonicAlpha.
- Plext: Shared context and reusable knowledge for AI coding agents, with an MCP server and skill integration.
- Weave: An AI content consultant that turns rough spoken ideas into structured, publishable content through agent workflows. Runner-up at Craft HackAIthon 2024.
- ChatGPT User Behavior Study (2025): Conducted a 750+ participant U.S. study using the kGrid AI Interviewer, converting open-ended responses into structured insights about real-world ChatGPT use.
- SpaceX IPO Retail Investor Study: Research involving 160 respondents, a paid report, a landing page, and an outreach campaign.
- Hands-on fine-tuning and post-training exploration: Human-in-the-loop and synthetic datasets, data quality, hyperparameters, evaluation, and iterative model improvement. This is experimental work, not a claim of a completed tool-calling fine-tuning product.

## Skills and working style
Python, JavaScript/TypeScript, Next.js/React, LLM agents, voice AI, RAG, embeddings, vector search, evaluation, fine-tuning, MongoDB, Elasticsearch, ETL, Cloudflare, Turbopuffer, and PostHog. Combines customer discovery, rapid prototyping, technical explanation, product ownership, and iteration from user feedback. Has also built growth, newsletter, and outreach systems.

## Education and awards
- M.S. Computer Science, Brandeis University, 2021–2023.
- B.A. Psychology, University at Buffalo, 2016–2018; summa cum laude.
- Three hackathon awards overall; the documented named project here is Weave, runner-up at Craft HackAIthon 2024.

## Outside work
Chess, soccer (Manchester United fan), running, lifting, and exploring coffee beans. Interested in how people think as well as how systems work.

## Find him
LinkedIn: https://www.linkedin.com/in/muthhukumar-malaiiyyappan-89aba054/
GitHub: https://github.com/Malai237

These are self-reported facts from Malai's resume, project notes, and personal site. Project involvement does not imply that he was the sole contributor. Metrics are historical, not current usage or revenue. Do not infer employment after September 2026, unlisted customers, funding, or outcomes.