Dandelion Engineering is a small research and technology company built around one human director working with AI agents. Its long-term aim is affordable technology that improves everyday people's lives. Its current work brings together practical engineering, research support, and experiments in human–AI collaboration.
The approach is still evolving. Early projects gave AI agents considerable responsibility for planning and carrying out research. Those experiences exposed limits in both the agents and the frameworks built around them. The direction now is closer collaboration: people bringing questions, judgment, and domain knowledge, with AI helping investigate, implement, test, and document the work.
The public repositories preserve that learning process, including unfinished projects, limitations, and approaches that did not work as hoped.
LLMs in Robotics: A Snapshot: An AI agent controls a simulated robot and uses persistent files to continue across sessions. Includes a short video and the human–AI conversations, code, tests, and results behind the experiment.
These repositories document early experiments with AI-assisted research. Their reports and artifacts remain available for inspection; their presence here is not an endorsement of every claim they contain.
- EEG Distillation: an exploratory analysis of reconstructing withheld EEG channels from a four-electrode input drawn from research-grade recordings. Reports and reproduction materials are available, with limits on what the experiment establishes.
- EEG to LFP: an exploratory study reporting a negative result for its specified cross-subject scalp-EEG decoding test, alongside an exploratory analysis that did not validate.
- Robot Structural Proprioception: ended on August 23, 2026 without answering its research question. The working record remains public.
- Hybrid Ground Truth Realism: ended on August 23, 2026 before the planned spike-sorting experiment was run. Its planning and corrections remain public.
These projects are not peer-reviewed. Reproducing reported numbers can help check an analysis, but does not by itself establish that the scientific question, methods, or interpretation are sound. The work needs domain-expert scrutiny before its conclusions are relied on.
- Collaboration Station 2.0: a general-purpose framework for keeping context, decisions, and work inspectable across long-running projects. It supports collaboration; it does not guarantee the quality of the work produced.
- Collaboration Station (original): the original general-purpose human–AI collaboration framework, now archived and preserved for reference.
- Collaboration Station Research 2.0: an experimental research-specific template. Dandelion Engineering considers it likely unsuitable for actual scientific research in its current form. It remains available as an experiment to learn from, not a recommended research workflow.
- Collaboration Station Research: the superseded first-generation research framework. Dandelion Engineering considers it likely unsuitable for actual scientific research in its current form. It is preserved as a record of the approach and its limitations.
Two smaller repositories demonstrate collaboration on scientific data pipelines: AccuSleePy and Vibration Batch Analysis. They are engineering demonstrations rather than evidence that AI can independently conduct reliable research.
Dandelion Engineering is run by Randy Crespo. Questions, critique, and conversations about research, engineering, workflow improvement, or collaboration are welcome.
This profile text was drafted by AI. Updated October 2026.