Work & Research
Work Experience
Startup under Entrepreneur First (VC), Anthropic for Startups, and Google Cloud for Startups.
Built a multiprocessing IDE in Rust that garnered 2k+ users and 300k visits in just 24 hours post-launch. Engineered a high-performance architecture capable of handling complex concurrent development workflows.
Worked with a team of 10 engineers to successfully recreate Google DeepMind's MuZero algorithm from the original research paper. Worked primarily on the neural networks team, implementing and optimizing the core model architecture.
Managed technical communications and bridged the gap between engineering and business teams. Secured significant funding and led critical negotiations for satellite equipment with major industry players like ESA, Kongsberg, and Nordic Semiconductor.
Conducted review of model code generation in Python and JavaScript. Contributed to the improvement of generative AI models by ensuring code quality, correctness, and adherence to best practices.
Designed the complete software architecture for an educational robot touring the EU. Delivered a comprehensive system blueprint that served as the foundation for the project's development and deployment.
Research Experience
Currently developing os12.com - a passion project research company focusing on AI native layer.
My work focuses on building and improving reinforcement learning systems, efficient computing models, and real-world applications of AI. I am passionate about reinforcement learning, efficient algorithms, and scalable automation, working across AI, robotics, and full-stack development.
Selected research and technical contributions are listed below.
Recursive Transformer Modules (RTMs)
Adi Singh, OS12 Research
Technical Draft. A unifying architecture for adaptive depth, parameter sharing, and auditable reasoning. RTMs invert the fixed-depth paradigm by recursively applying a shared reasoning block to a persistent latent workspace.
Async + Parallel LLM Coding Agents
Adi Singh, OS12 Research
Technical Draft. A framework for asynchronous and parallel execution in LLM coding agents. Investigates parallel speculative generation and dynamic scheduling to reduce latency while preserving solution quality.
Benchmarking Classic and Modern RL Algorithms: REINFORCE, DQN, and Tabular Q-Learning on CartPole-v1
Adi Singh, Jan Christian Meyer
PhD Research Collaboration, Department of Computer Science, In Progress (2025)
AI-Based Matchmaking (CatMatch): A Novel Convolutional Neural Network Approach to Preference Learning
Adi Singh, Cogito NTNU
AI Tinder training algorithm using CNN, Proceedings of the Student Project at Cogito (2023)
Cost-Efficient Treatment Methods for Hydrocephalus: Data Analysis and Optimization
Adi Singh
MIT Biogen Community Lab, Journal of Medical Research & Innovation (2022)
Education
Master's/PhD Level:
Computer Vision and Deep Learning, Methods in Artificial Intelligence, Advanced Parallel Computing, Network Programming & Security
Bachelor's Level:
Engineering Statistics, Linear Algebra, Procedural & Object-Oriented Programming, Control Theory (1+2)
Awards & Recognition
Meta Generative AI Hackathon — 4th Place, 2025
Jane Street Estimathon — 1st Place, 2024
Start Code Hackathon — 2nd Place, 2024
TripleTex Hackathon — Finalist, 2024
Building 15,000+ Follower ML Channel — 2024
EECS Representative — 2024-Present
Advanced Math Scholar — Virginia Richmond
High School Representative — 2021