Portfolio home

ADAM AI app walkthrough

A four-minute tour of the app, recorded live from a local build on public CDISC pilot data (pharmaverse). It covers study setup, the pair programmer, traceability, QC tools, submission checks, and a full agent run that ends with sign-off by a second user.

Proof of concept, local build

Chapters

  1. Sign in and the workspace
  2. Create a study, choose CDISC versions, upload the ADaM spec (28 variables, 7 categories)
  3. Pair programmer: variables by category with their derivation rules
  4. Traceability matrix and lineage graph
  5. SDTM test data, dataset comparison and code comparison
  6. Submission readiness checks
  7. Agent run: static checks, R run on 306 subjects, comparison with the reference (attempt 1 fails, attempt 2 matches 28/28)
  8. Independent sign-off by a second user, and the audit log check

No AI model is called in this recording. In the agent run, two recorded programs stand in for the model (replay mode, the same one the public benchmark uses); the first carries two planted bugs. The static checks, the R run, the comparison with the reference, the sign-off rules and the audit log all run live. The pipeline is also shown in the one-minute pipeline demo. Code: github.com/grprakasam/adam-ai.