Haney GPT & Haney CLI Coding Agent

#python #ai #productivity #opensource

I Built a 537M Parameter GPT Model and an AI Coding Agent as an Independent Learner

Over the past few years, I have spent much of my time learning and building in the AI space. While my path into AI engineering has been unconventional, one thing has remained constant: curiosity.

That curiosity led me to build two projects that taught me more than any course ever could:

🐱 Haney GPT

and

🐱 Haney CLI

Haney GPT

Haney GPT started as an attempt to understand how large language models work under the hood.

Rather than treating LLMs as black boxes, I wanted to learn about tokenization, transformers, training pipelines, datasets, optimization, and inference by building things myself.

Over time, the project evolved into a GPT-style model that I scaled to 537 million parameters.

The process taught me lessons about:

Data preparation
Training stability
GPU limitations
Evaluation
Model deployment
LLM architecture

Most importantly, it gave me a much deeper appreciation for the engineering challenges involved in modern AI systems.

Haney CLI

After working with language models, I became interested in AI agents and developer tooling.

That led me to build Haney CLI, an AI-powered command-line assistant designed for developers.

Install:

pip install haney

Haney CLI follows a BYOK (Bring Your Own Key) approach and currently supports multiple model providers, allowing developers to use the models that best fit their workflow.

Current capabilities include:

6 AI providers
10 MCP integrations
Plan Mode
Edit Mode
Developer-friendly CLI workflows
Extensible architecture

The goal is not to replace developers but to create a practical AI companion that can help with coding, research, planning, and automation tasks.

Lessons Learned

Building AI systems independently taught me that the hardest problems are rarely technical.

The biggest challenges are:

Understanding user needs
Creating good developer experiences
Handling edge cases
Maintaining simplicity while adding power

Every feature seems easy until real users start using it.

Building in Public

I continue to document my learning journey through projects, experiments, and open-source work.

Projects:

🐱 Haney CLI
https://codehaney.dev

📚 Sakthi Wiki
https://sakthi.wiki

Sakthi Wiki is an LLM-powered wiki where I document concepts, experiments, and learnings from AI engineering, machine learning, and software development.

What's Next?

I am continuing to explore: