Could a team build AI adverse event grading like Burna AI in a weekend?

A weekend produces a classifier, or one model call that returns a grade. Burna AI's pipeline spans twelve agent categories: extraction, verification, resolution, standardization, taxonomic matching, differential analysis, temporal tracking, documentation integrity, evidence synthesis, causal attribution, confidence calibration and regulatory encoding, with each agent's output bounding every agent after it. Around the engine sit both CTCAE versions, WHO-UMC and Kramer attribution, MedDRA coding, blind grading and adjudication, 21 CFR Part 11 signatures, an isolated data plane for protected health information, and SMART on FHIR integration. Two patents filed on the cascading constraint pipeline.