# Kunal Verma

AI Engineer · Founder at InfraHive.ai

Location: Jaipur, India

AI Engineer and Founder at InfraHive.ai — Data/Context Layer for Enterprise AI deployed at Fortune 500 companies, with built-in data connectors. I've been building since 2020, going through multiple pivots, failures, and learnings to arrive at what InfraHive is today. Curious person with experience across AI infrastructure, NLP, and enterprise data, might contribute to AGI someday

- [GitHub](https://github.com/hereiskunalverma)
- [LinkedIn](https://www.linkedin.com/in/kunalverma19/)
- [Twitter/X](https://x.com/19kunalverma)
- [Medium](https://medium.com/@hereiskunalverma)

## Timeline

### Founder — InfraHive.ai

*Present*

Data platform for AI labs and enterprise teams — combining data collection, expert annotation, and evaluation infrastructure to help teams train, test, and improve advanced AI models. The same layer turns company knowledge into Playbooks for AI Agents, run on custom-built and open-source agent harnesses — similar to what Claude Code is for coding, built for domain-specific use cases.

- Customer stories:
  - Evaluation — Evaluation loops on a mid-market SaaS company's support corpus cut Tier-1 ticket resolution from 4 hours to 12 minutes.
  - Annotation — Expert-labelled ticket data and continuous evaluation keep a B2B manufacturing platform's 24/7 AI support layer handling 30+ tickets/week.
  - Data Pipelines — A $28M DTC brand moved off NetSuite onto InfraHive's MetricFlow, unifying financial data to close books 5x faster.
  - Data Pipelines — Godrej Properties migrated humungous sales data with AI-powered pipelines after multiple vendor failures.
  - Data Pipelines — $5B Ecommerce giant transforms BigQuery analytics with InfraHive's ETL layer for natural language querying.

### Techstars Tokyo 2024 — Techstars

*2024*

Alumni of first Tokyo Batch

### Aivinya AI — (pre-InfraHive)

*2020 – 2023*

2 pivots: B2C AI Characters → B2B AI Characters for agencies → InfraHive.ai.

### Google Kick Start 2022 — Google

*2022*

Global online coding competition — placed among top 500 worldwide.

### B.Tech in Computer Science & Engineering — IIIT Kota

*2020 – 2024*

Indian Institute of Information Technology, Kota.

## Projects & Research

### MuMo-RAG

Multi-modal Retrieval-Augmented Generation for medical report generation. Outperforms existing solutions on IU X-Ray and MIMIC-CXR benchmarks.

Tags: RAG, Multi-modal, Medical AI

### Compact Intelligence

Research on smaller LLMs (Llama, Mistral) vs GPT-3.5/GPT-4 for medical QA — cost-effective AI in high-stakes domains.

Tags: LLMs, Medical AI, Research

### [Neural Style Transfer (TF.js)](https://github.com/hereiskunalverma/neural-style-transfer-tfjs)

Neural style transfer running directly in the browser with TensorFlow.js.

Tags: TensorFlow.js, Browser ML

### [ChatBot using Reformer](https://github.com/Dev-is-here/ML-DL-Major-Projects)

Pre-ChatGPT era chatbot built with the Reformer (efficient Transformer) architecture.

Tags: Transformers, NLP

### [Introduction to NLP](https://github.com/hereiskunalverma/introduction_to_NLP)

Educational resources and implementations for Natural Language Processing.

Tags: NLP, Education

### [Energy Efficiency of Buildings](https://github.com/hereiskunalverma/Random_Repository/tree/main/Energy%20Efficiency%20of%20Building)

ML model predicting heating/cooling loads from commercial building parameters.

Tags: Machine Learning, Regression

## Skills

- LLMs & Finetuning
- Machine Learning
- NLP
- Embeddings & Reranking
- Python
- React / Next.js

## Certifications

- Machine Learning
- Deep Learning Specialization
- AdMat (Advanced Matlab and Its Applications)
- Digital Identity Hack 2022 Winner
- Innovation Challenge Hackathon Finalist: Team REXIS

## Hobbies

- Poetry
- Reading — literature, biographies, math, research papers
- Mohammad Rafi's top 0.01% listener on YouTube
- Writing -- barely
- Handball (National Level-u19 Asian Handball Federation) & recently, Pickleball -- just started
- Art (District Level Winner) -- rusty