Compliance monitoring · Production

Detecting missing material facts in healthcare advertising using ApertureDB, Unstructured, and OpenAI

The problem

Healthcare marketing content — especially social media posts promoting prescription drugs — frequently omits FDA-required material facts such as contraindications and safety limitations because informal channels like influencer partnerships often skip formal review processes.

First attempt

Kim Kardashian's post promoting Diclegis omitted the drug's key contraindication — that it had not been studied in women with hyperemesis gravidarum — resulting in an FDA warning letter, illustrating that informal content channels lack systematic compliance checks.

Workflow diagram · grounded in source
1
Marketing document provided
trigger
“marketing_claim_image: "data/KimKardashianAd.png"”
2
Clinical PDF ingested to ApertureDB
integration
“We pull text and images from clinical PDFs, split the content into smaller pieces, embed them, and store everything in ApertureDB. Since these are medical documents, we use a fine-tuned embedding model optimized for medical benchmarks. T…”
3
LLM detects omissions
ai_action
“We prompt an LLM to detect potential omissions. The model identifies missing details, including known limitations, contraindications, and required evidence. The prompt is designed to reduce the risk of hallucinated outputs.”
4
Vector similarity search against clinical data
ai_action
“After an LLM flags potential omissions, we verify them against trusted clinical PDFs. Using ApertureDB, we run similarity searches to cross-check the facts and confirm their accuracy.”
5
LLM validates omission consistency
validation
“Decide if the documents support the observation or not: - If yes, return status "Omission" - If not, return status "Fine"”
6
Flagged omissions displayed
output
“Category : omitted contraindications Observation : There is no mention of specific health conditions or demographic populations for whom the drug might be dangerous or ineffective. → Status: Omission - The provided documents highlight im…”
Reported outcome

The pipeline identifies omissions in healthcare marketing content — including the specific contraindication cited in the FDA's warning letter — and automates what previously required manual review, saving time and building credibility.

Reported metrics
Time saved by automationsaves time
Omission detection effectivenessidentifies various omissions, including the one mentioned in the FDA's warning letter
Reported stack
ApertureDBUnstructuredOpenAIeasyocrsentence_transformersCLIP
Source
https://mlops.community/blog/how-multimodal-vector-databases-are-transforming-challenges-across-industries
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The pipeline identifies omissions in healthcare marketing content — including the specific contraindication cited in the FDA's warning letter — and automates what previously required manual review, saving time and bui…

What tools did this team use?

ApertureDB, Unstructured, OpenAI, easyocr, sentence_transformers, CLIP.

What results were reported?

Time saved by automation: saves time; Omission detection effectiveness: identifies various omissions, including the one mentioned in the FDA's warning letter (source-reported, not independently verified).

What failed first in this deployment?

Kim Kardashian's post promoting Diclegis omitted the drug's key contraindication — that it had not been studied in women with hyperemesis gravidarum — resulting in an FDA warning letter, illustrating that informal con…

How is this compliance monitoring AI workflow structured?

Marketing document provided → Clinical PDF ingested to ApertureDB → LLM detects omissions → Vector similarity search against clinical data → LLM validates omission consistency → Flagged omissions displayed.

Related compliance monitoring cases