Akshay Iyer

Founder, Flomi AI

About Me

I'm super driven about creating something very impactful using AI. Won't stop until I get there.

Working on building the best way to transfer the specific knowledge of enterprises to AI agents.

Here are some of my test scores:
• IIT-JEE (India's toughest engineering entrance exam) : All India Rank 225 among 1.3 million students (top 0.01 percentile).
• Among the national top 1% of test takers in:
- Indian National Physics Olympiad 2019
- Indian National Chemistry Olympiad 2019
- Indian National Astronomy Olympiad 2019
• KVPY fellowship recipient (Indian Government): All India Rank 109.
• Current Columbia University CGPA: 4.04/4.

Here are my publications:
• AI research paper to mitigate training data biases. AAAI conference paper: Read here
• Akshay Iyer*, Kanad Pardeshi*, Nikunj Goyal*, Pranamya Kulkarni*, Abhilasha Sancheti et al., "A method to generate image descriptions with varying levels of detail", filed patent (USPTO), Adobe reference: P11617-US.

Education & Experience

Columbia University in the City of New York

MS CS | Aug 2023 - Dec 2024 | New York, United States

GPA: 4.0825/4.0
Relevant Coursework: Generative AI for Code, Cloud Computing, Engineering Software as a Service, Blockchain and Web3.0, Databases, Computation and the brain, Principles of Innovation, UI Design

Here's some stuff I did :

Zerocurve AI: Built a tool that helps people learn to use complicated-UI software like the Epic EHR software.
Inntwine: Built a networking and idea sharing platform for Columbia students.
Software-as-a-Service Developer: Built a SaaS platform for food makers, using Ruby on Rails, following agile development.
Blockchain Developer: Built an Ethereum-based FCFS freelancing platform with automatic job verification and payment.
AWS Cloud Application: Built a microservice-based courseworks add-on with calendar integration for Columbia students.
Brain Simulator: Developed a simulator to analyze transition dynamics of neuron assemblies in spiking networks.
Teaching Assistant (COMS 4444): Created game simulators, supported 30+ students trying to beat it in real time.
Teaching Assistant (COMS 4460): Guided 20+ students on how to best launch their tech venture.

Indian Institute of Technology, Bombay

BTech. Electrical Engineering and Computer Science | Jul 2019 - Aug 2023 | Mumbai, India

GPA: 9.21/10.0
Relevant Coursework: Technology Venture Creation, Data Structures and Algorithms, Analysis of Algorithms, Programming for Data Science, Introduction to ML, Advanced ML, Intelligent Agents and RL, Deep Learning for NLP, Optimization in ML, Digital Image Processing

Here's some stuff I did :

Data partnerships in Indian banks: Worked on a B2B to greatly simplify bank digital partnerships (customer data exchange with third parties).
AI Researcher: Focused on counterfactual estimation for continuous treatment datasets. Published at AAAI-24.
Neuromorphic Engineering Research: Developed speech recognition using Liquid State Machines (LSM) and explored non-backpropagation training rules to improve performance.
GAN Perception: Trained GANs on different input image datasets and used CLIP to gauge their understanding of subjective concepts like 'beautiful'.
Institute Student Mentor: Mentored 12 freshmen on academics and extracurriculars.
Teaching Assistant (MA207): Conducted tutorials on advanced differential equations for 40+ students.

Some things I've built

The IDE for non technical users

Cursor is for technical people. Lovable and Bolt are for frontend and only very simple websites. Non technical people want to build more, and my tool enables just that.

Karen.ai - Stress Test for Customer Support

This is Karen.ai, a website to stress test your customer support employees by being a Karen.

Flomi AI - Enterprise Documentation Tool

A tool to make documenting the specific knowledge in enterprises much easier, things that usually go undocumented.

AI agent marketplace

A way to instruct AI agents to do workflows using examples. People have specific knowledge, and examples seemed to be the best way to capture that.

Mirror AI - Your AI Digital Persona

A platform to create your own digital persona. It's an AI that talks like you and that you can feed your information to in an authenticated way. Allow your friends to chat with it and embed it in your own website as an API. Built for the YC + Supabase hackathon (Got a special mention).

Agent-based automation platform

Current automation platforms are almost like visual coding (blocks and flowcharts), which is too complex for the average HR/accounting person to learn and do quickly. I'm trying to build a much more intuitive automation platform for enterprises using Agentic AI and LLM Code generation.

Unit testing for LLM prompts

Created a library that captures LLM input prompts you put in during development and uses an LLM to generate variations of these input-output pairs. This enables developers to evaluate the reliability of their system prompts in different ways. Entrepreneur first hackathon project.

AI agents to automate accounts

I used AI agents to automate tedious accounting tasks like invoice data entry and ledger scrutiny. Used Langchain and LangGraph. Human in the loop for important actions.

Big Tech Experience

Adobe

Machine Learning Researcher - May 2022 - Jul 2022

As a Machine Learning Researcher at Adobe, I was responsible for extensive literature survey in multimodal media and developing a controllable image description generation system.

Analyzed over 30 research papers in multimodal media to formulate a research problem.
Utilized dense captioning through Faster R-CNNs and BERT transformers for controllable image captioning.
Developed and submitted a patent for the first controllable captioning algorithm using a continuous value as input.
Presented slide decks and demonstrated the solution to senior executives at Adobe.
Created a website to deploy ML models and provided real-time demos.

Resulting Publication: Akshay Iyer et al., "A method to generate image descriptions with varying levels of detail", Adobe Reference: P11617-US

Python Natural Language Processing (NLP) Computer Vision Deep Learning Machine Learning

Samsung Semiconductor

Digital Hardware Intern - May 2021 - Jul 2021

As a Digital Hardware Intern, I implemented data integrity checks for the NVMe interface in Solid State Drives (SSDs).

Analyzed the NVMe interface, NVMe IPs, DDR memory, and the axi4 interface.
Verified the 8-bit protection information appended to stored data to ensure data integrity in SSDs.

VHDL NVM Express (NVMe)

Cogno AI

Machine Learning Intern - Dec 2020 - Jan 2021

At Cogno AI, I built a customer service chatbot using NLP and deep learning techniques.

Processed data using language translation tools and NLP techniques such as lemmatization and stemming.
Implemented a bi-directional LSTM neural network to classify queries into predefined sets.
Achieved a classification accuracy of over 90% across a multilingual dataset.

Python Natural Language Processing (NLP) Machine Learning