By Swapnil
Curiosity,
made useful.
I’m interested in how AI systems work, where they fail, and what it takes to make them useful. Model Fieldnotes is where I turn those questions into tools and explanations.
A place to work things out
The work here connects three things: open-source applications you can use, research you can inspect, and practical explanations you can challenge.
Agent Explainer is the first project: an interactive lab for understanding agent failures. The fieldnotes explore the surrounding decisions, from retry behavior and model evaluation to cost and governance.
This is an independent personal project. The writing represents the perspective of Model Fieldnotes, and does not speak for an employer, vendor, or research institution.
Editorial principles
- Follow the evidence. Link to original papers and official documentation. Distinguish source findings from interpretation.
- Name the limits. A simulation teaches a mechanism. A measured experiment supports a claim within its tested conditions. Neither is a universal model ranking.
- Show the work. Reproduced experiments should include their setup, model versions, prompts, dates, and limitations.
- Keep the dates visible. Pricing, model capabilities, and regulation change. Articles show their publication and source-review dates.
- Correct in the open. Substantive corrections receive an update date and explanation. The source history remains available.
- Be clear about assistance. AI assists research, drafting, and software development here. Assistance is not independent validation; readers can inspect the cited evidence.
Follow or contribute
Read new writing through the RSS feed, follow swapupg on GitHub, or suggest a correction or a question. Contributions that make an explanation clearer are as welcome as code.