Sabareesh

AI Researcher and Founder

Research on language models that keep learning after deployment: continual learning, mid-training, post-training, agents, and memory

[email protected]
sabareesh.com
linkedin.com/in/sabaree
github.com/banyan-god

Profile

I research how language models keep learning after pretraining: domain mid-training, supervised and reinforcement-learning post-training, test-time adaptation, memory, and agents that manage their own context. I also build the research software that makes those experiments repeatable.

I come to this as an operator. I spent 13 years building and running production software in enterprise and fintech, including four years as CTO, where I was accountable for engineering, cloud infrastructure, security, and AI.

Research

Founder
2026 – Present

Focus. Models are usually trained once and then frozen. My work asks how they can keep adapting after deployment, remember across sessions, and delegate work to agents with their own context and tools.

Research agenda

  • Continual learning and test-time training: updating a model’s weights on the sequence it is reading, and understanding what that does to long-context behavior.
  • Mid-training and post-training: continued pretraining, supervised fine-tuning, and reinforcement learning to move a general model into a domain and then improve its decisions.
  • Agents and memory: recursive and multi-agent systems that manage their own context, with semantic memory for long-running work.
  • Research automation: autonomous software that runs, tracks, and learns from experiments.

Looking for: research collaborations and business opportunities across Asia and North America.

Operating Experience

GuidedChoice — Reno, NV

2017 – August 2026

Chief Technology Officer
2022 – August 2026

  • Led technology strategy, software architecture, engineering, cloud infrastructure, information security, and AI development and integration for digital financial products, including 3nickels.
  • Directed application development and technical delivery; maintained 99.9%+ uptime and led the migration from Docker Swarm to Kubernetes to improve scalability and resilience.

Product Manager, Architect and Lead Developer
2017 – 2022

  • Architected a microservices platform, modernized legacy applications, managed an Oracle-to-SQL Server migration, and mentored engineers in Spring Cloud and service architecture.

ARS National Service — Escondido, CA

Enterprise Architect and Senior Developer
August 2013 – September 2017

  • Introduced Docker and Docker Swarm across data centers; implemented WSO2 API Manager and built CRM APIs, vendor integrations, ETL tools, and document management systems.

California State University San Marcos — San Marcos, CA

iOS Developer
Sep 2012 – Mar 2013

National Science Foundation web project
Sep 2011 – Jun 2012

Technical Foundation

  • Model training: mid-training, continued pretraining, supervised fine-tuning, reinforcement learning (GRPO, DAPO, PPO), meta-learning, test-time training, PyTorch, torchtune, torchtitan, torchao, TRL, NeMo-RL
  • Systems: CUDA, FSDP/FSDP2, DDP, vLLM, llama.cpp, Ray, Hugging Face, Weights & Biases
  • Agents: recursive language models, multi-agent orchestration, MCP, context compaction, semantic memory, observability
  • Software and infrastructure: Python, Java, TypeScript, C/C++, Swift, FastAPI, Spring Boot/Cloud, Docker, Kubernetes, Azure, AWS, SQL Server, Kafka

Education

Master of Science in Computer Science — California State University, San Marcos, 2013

Bachelor of Technology in Information Technology — Easwari Engineering College, Anna University, Chennai, India, 2010