Live demo· CDISC pilot / synthetic data only· Anonymous quota: 5 runs/day· Generated code is an ungated draft scaffold requiring human review

Live demo / governed TFL platform

Try the platform live.

The demos elsewhere on this site are precomputed or run in your browser. This page is different: it submits real jobs over a versioned JSON contract to a governed API (app.jaimeyan.com), where a containerized R worker executes them and returns hash-pinned artifacts. Same engineering standard as the platform concept demo, but the round trip is real.

No account, no API key, no upload path: anonymous callers get 5 runs per day per IP, and only built-in synthetic / CDISC pilot data can be referenced. Every artifact comes with a SHA-256 so you can verify what you downloaded is what the worker wrote.

Checking remaining demo quota…

Three contracts, one queue

Submit a job. Watch it run. Keep the evidence.

Each card below posts a contract-tagged JSON body to POST /v1/jobs, polls until the worker finishes, then lists the resulting artifacts. The flagship card runs a synthetic study review where an independent QC gate can block the report while the evidence stays available. The request body is shown verbatim — nothing is hidden between this page and the API.

01 / synthetic-report/0.1

Synthetic study review

Run a fixed synthetic study — 12 fictional participants, a safety population of 9, and 3 summary tables. R builds ADSL and the summaries from the inputs while Python independently recomputes reference results. Failed or unverified QC blocks the report while keeping the evidence available for review.

Request body (JSON) 
 
  
  
 
 
  
 
02 / adam-build/0.1
 
Build an ADaM dataset
 

Rules backend derives an ADaM dataset from the built-in CDISC pilot SDTM, with a
							generated admiral program and a hash-chained run log as first-class artifacts.

 
   
 
 Request body (JSON) 
 
  
  
  
 
03 / figure-spec/0.1
 
Render a figure
 

A declarative figure spec — kind, column mapping, output formats — rendered by the
							worker from a synthetic ADaM-style teaching bundle. The SVG renders inline below.

 
   
 
 Request body (JSON) 
 
  
  
 
  
 
 
Also live / package governance
 
R package validation gateway
 

Tier-1 screening for R packages under a two-tier validation strategy: riskmetric
							risk scores plus deterministic evidence produce a config-driven
							GO / GO-WITH-CONDITIONS / NO-GO decision, an evidence-linked report, and a
							hash-chained audit record. Submit a CRAN package name or a public GitHub URL —
							assessments count against the same anonymous daily quota. Tier-2 full validation
							(sandboxed R CMD check, testthat/covr evidence, dual-signer validation package)
							is available as a managed service —
get in touch.

 Open the validation demo → 
 
 
 
  
 
 
Boundaries, stated plainly
 

What this demo is — and is not.

 
 
 
Real pipeline, synthetic data
 

Jobs execute on the same gateway, queue, and sandboxed R worker described in the
platform architecture.
							Only built-in synthetic and CDISC pilot data are reachable; there is no upload path
							for anonymous callers, so nothing you do here can touch real study data.

 
 
 
Quota as a governance feature
 

Anonymous access is capped at 5 runs per day per IP, with a platform-wide daily
							circuit breaker behind it. Exceeding the limit returns 429 with a
Retry-After — the same fail-closed posture the platform applies
							everywhere else.

 
 
 
Drafts, not deliverables
 

Generated datasets, figures, and programs are teaching artifacts. Generated code is
							an ungated draft scaffold: it requires human review, independent QC, and study-level
							validation before it could support any regulated use.

 
 
 
 
What this page deliberately is not
 
     
  • It is not a production analysis environment. Figures and datasets are rendered from synthetic fixtures and are not suitable for any submission or decision.
    •  
  • It does not accept your data. Uploaded specs and LLM-backed generation require an invite token and are not exposed here.
    •  
  • It makes no LLM calls: all three contracts run deterministic worker code. The generated draft still requires human review — determinism is not correctness.
    •  
  • The synthetic study review is an engineering demonstration of a gated evidence flow — it does not establish submission-ready ADaM compliance or independent human sign-off.
    •