Aleksei Romanov

Ex-Apple GenAI engineer, building an LLM post-training platform at g factor technologies

Joined on Sep 21, 2026 https://www.g-ftech.com

Aleksei Romanov for g factor

Oct 10

Teaching a 27B Model to Write Trading Alphas: 101 Formulas, 12 Rewards and One Unseen Year

#machinelearning #llm #reinforcementlearning #ai

17 min read

Want to connect with Aleksei Romanov?

Create an account to connect with Aleksei Romanov. You can also sign in below to proceed if you already have an account.

Create Account

Already have an account? Sign in

Aleksei Romanov for g factor

Oct 9

Multi-Reward RL, Part 3: GDPO + CISPO + REPO-R at 27B, and the Advantage Floor That Stopped Learning

#machinelearning #llm #reinforcementlearning #ai

12 min read

Aleksei Romanov for g factor

Oct 7

Pushing the Limits: Extreme Inference Speedup of Qwen 3.8 27B on NVIDIA B300 (100 to 10k+ tok/s)

#llm #machinelearning #performance #ai

16 min read

Aleksei Romanov for g factor

Oct 6

Multi-Reward RL, Part 2: Benchmarking GRPO, DAPO, and CISPO on Unseen Tasks

#machinelearning #llm #reinforcementlearning #ai

11 min read

Aleksei Romanov for g factor

Oct 5

The Verifier Design Playbook: How to Build RLVR Gyms That Models Can't Game

#ai #machinelearning #llm #reinforcementlearning

11 min read

Aleksei Romanov for g factor

Oct 4

Shopify, Intercom, Ramp: How Open Models Beat Frontier APIs in Production

#ai #llm #machinelearning #agents

14 min read

Aleksei Romanov for g factor

Oct 3

Long-Horizon Agents: Why Multi-Turn Reasoning Breaks and the Practical Training Tricks That Fix It

#ai #llm #machinelearning #agents

12 min read

Aleksei Romanov for g factor

Oct 2

We Tried ISO-AdamW. AdamW Kept Its Job.

#machinelearning #llm #deeplearning #ai

11 min read

Aleksei Romanov for g factor

Oct 1

Benchmarking Qwen 3.8 27B Across Inference Providers: Together, Fireworks, Doubleword, and g factor

#llm #performance #gpu #machinelearning

14 min read

Aleksei Romanov for g factor

Sep 30

The Limits of AI: Induction, Deduction, and Why Models Can't Jump

#discuss #ai #machinelearning #llm

11 min read

Aleksei Romanov for g factor

Sep 30

Physical AI: Why the Next Big Frontier Is Giving Software Agents Hands

#ai #robotics #machinelearning #llm

10 min read

Aleksei Romanov for g factor

Sep 29

Distributed Training & Inference: From CPUs and GPUs to a Cluster

#machinelearning #pytorch #gpu #distributedsystems

14 min read

Aleksei Romanov for g factor

Sep 28

From RLHF to RLVR: The Evolution of Reward Signals and the Battle Against Reward Hacking

#machinelearning #reinforcementlearning #llm #security

9 min read

Aleksei Romanov for g factor

Sep 27

AsyncGRPO: Eliminating GPU Idle Bubbles in Environment-Heavy RL Post-Training

#machinelearning #reinforcementlearning #llm #gpu

9 min read

Aleksei Romanov for g factor

Sep 27

Owning vs. Renting Intelligence: Why Enterprises Are Building Sovereign AI

#ai #llm #machinelearning #mlops

9 min read

Aleksei Romanov for g factor

Sep 26

LoRA & DoRA: The Math, Memory, and Trade-offs

#machinelearning #ai #llm #deeplearning

9 min read

Aleksei Romanov for g factor

Sep 25

Multi-Reward Reinforcement Learning for LLM Agents: Comparing PPO, GRPO, DAPO, and GDPO

#machinelearning #ai #llm #reinforcementlearning

8 min read

Aleksei Romanov for g factor

Sep 24

SFT vs. RL: What Changes Inside the Model?

#machinelearning #ai #llm #deeplearning

6 min read

Aleksei Romanov for g factor

Sep 23

Latent-GRPO: Reinforcement Learning in Continuous Thought Space

#grpo #machinelearning #ai #llm

9 min read

Aleksei Romanov for g factor

Sep 21

High-Throughput LLM Inference & Training: A Deep Dive into vLLM

#ai #machinelearning #python #gpu

9 min read

DEV Community

Aleksei Romanov

Badges

2 Week Community Wellness Streak

1 Week Community Wellness Streak

Writing Debut

Organizations

g factor

Teaching a 27B Model to Write Trading Alphas: 101 Formulas, 12 Rewards and One Unseen Year

Want to connect with Aleksei Romanov?

Multi-Reward RL, Part 3: GDPO + CISPO + REPO-R at 27B, and the Advantage Floor That Stopped Learning

Pushing the Limits: Extreme Inference Speedup of Qwen 3.8 27B on NVIDIA B300 (100 to 10k+ tok/s)

Multi-Reward RL, Part 2: Benchmarking GRPO, DAPO, and CISPO on Unseen Tasks

The Verifier Design Playbook: How to Build RLVR Gyms That Models Can't Game

Shopify, Intercom, Ramp: How Open Models Beat Frontier APIs in Production

Long-Horizon Agents: Why Multi-Turn Reasoning Breaks and the Practical Training Tricks That Fix It

We Tried ISO-AdamW. AdamW Kept Its Job.

Benchmarking Qwen 3.8 27B Across Inference Providers: Together, Fireworks, Doubleword, and g factor

The Limits of AI: Induction, Deduction, and Why Models Can't Jump

Physical AI: Why the Next Big Frontier Is Giving Software Agents Hands

Distributed Training & Inference: From CPUs and GPUs to a Cluster

From RLHF to RLVR: The Evolution of Reward Signals and the Battle Against Reward Hacking

AsyncGRPO: Eliminating GPU Idle Bubbles in Environment-Heavy RL Post-Training

Owning vs. Renting Intelligence: Why Enterprises Are Building Sovereign AI

LoRA & DoRA: The Math, Memory, and Trade-offs

Multi-Reward Reinforcement Learning for LLM Agents: Comparing PPO, GRPO, DAPO, and GDPO

SFT vs. RL: What Changes Inside the Model?

Latent-GRPO: Reinforcement Learning in Continuous Thought Space

High-Throughput LLM Inference & Training: A Deep Dive into vLLM