study-skills2026
Skills for studying with an AI agent: it researches the topic, builds a plan, writes study pages, assesses you and schedules reviews. It works for studying on your own or for turning your classes into study pages.
About
Systems Engineering student at Universidad del Quindío (2nd semester), in Armenia, Colombia. I'm learning to build AI agents in Python.
I work by defining the specifications and the process of each project, and I implement it by directing AI agents. I'm constantly developing my directing skills and my understanding of code, to sharpen my judgement and make better decisions. I also enjoy writing code by hand, even though it's becoming less and less efficient, so I do it more as a hobby and out of passion.
Now
- harness, my agent written in Python. It already chats and compacts its context; now I’m adding tools.
- Next on my roadmap: evals and tests, my own MCP server, connecting harness to it, rebuilding it with LangGraph and deploying it on AWS.
- Local models: I run and benchmark them on my homelab, which has an RTX 3050 with 6 GB.
Projects
Vaecos tracking2026
The logistics app I built for my family's store, which handles around 30 orders a day. It tracks shipment status and generates shipping labels; about 15 in every 20 are created automatically.
recipes2026
Measured recipes for running open image and text models on modest hardware: the exact command, the winning settings and the numbers behind them. I measure them on an RTX 3050 with 6 GB and on a CPU-only laptop.
visor2026
An app for studying with AI, in the spirit of Obsidian: it shows folders of HTML pages and Markdown notes. You highlight and comment, and your agent reads it and replies on the same page. With a simple note it uses about half the memory of Obsidian.
harness2026
A terminal agent in Python that talks to any OpenAI-compatible server. It has streaming, token counting and automatic compaction of the conversation once it passes 85% of the context.
Technologies
- I use
- Python, Java, Docker, llama.cpp, Ollama, Git, Linux
- Learning
- agents and tool calling, MCP, FastAPI, LangGraph