Everyone's buzzing about "AI literature review" tools. Most are just wrappers on the same LLMs with a fancy UI. For a pharma team of five, you don't need the hype—you need structured, traceable outputs you can audit.
Elicit is fine for the initial scrape. But you'll drown in PDFs. The real work is the old-school pipeline: get the metadata and PDFs out of Elicit, land them in a data lake, and structure the findings yourself. Use a simple orchestration to pull from Elicit's API, then load to your warehouse. Here's a basic pattern:
```sql
-- Create a staging table for the papers Elicit returns
CREATE TABLE stg_elicit_papers (
paper_id VARCHAR,
title VARCHAR,
authors VARCHAR,
abstract TEXT,
pdf_url VARCHAR,
elicited_metrics JSONB,
load_timestamp TIMESTAMP
);
```
Then use dbt to model it cleanly. Don't let the tool do your "insights." Write your own SQL to find patterns, gaps, contradictions. Tools come and go. A solid fact table in Snowflake or BigQuery is forever.
SQL is enough