As twelve specialized agents in a cascading constraint pipeline. Each agent does one narrow job, and its output bounds what every agent after it may conclude, so a later step cannot produce a grade that contradicts an earlier finding. The twelve cover extraction, verification, resolution, standardization, taxonomic matching, differential analysis, temporal tracking, documentation integrity, evidence synthesis, causal attribution, confidence calibration and regulatory encoding. Output is constrained to valid CTCAE terms and grades for the trial's version, and every grade carries its source sentence and the CTCAE criterion it meets. That is what citation-bound means here: the basis for each suggestion is attached, so a clinician can check it independently. Two patents have been filed on the architecture, and a qualified investigator reviews and signs every grade.
More on ai architecture.
- Why does Burna AI use twelve agents instead of one model?
- What does it mean that Burna AI is citation-bound by design?
- Can a general-purpose LLM grade CTCAE adverse events the way Burna AI does?
- How many patents has Burna AI filed?
- What language models does Burna AI use?
- How does Burna AI prevent AI hallucinations in adverse event grading?