Stop Hallucinations. Ground Every Health Answer.

What you get

• Hybrid retrieval: BM25 + dense FAISS with reciprocal rank fusion
• Quality tiers: every source carries A/B/C/D rating + numeric quality score
• Citation IDs: PMIDs, DOIs, NCT IDs, MeSH terms in structured fields
• Source classes: Cochrane, meta-analysis, RCT, guideline, regulatory, ODS factsheet
• Anti-hallucination prompt block: a vetted system prompt you can compose into your own LLM
• Stable response schema: contracted at /v1/* with deprecation notice (P1 roadmap)

Reference architecture

1. User question arrives at your application.
2. Call GET /api/sources with a topic vertical and min_score threshold to pull the top-N evidence rows.
3. Build your LLM context using title, quality_tier, pmid, and the abstract excerpt.
4. Generate the answer with your model, instructing it to cite via [PMID:XXXXXXXX].
5. Optional: call GET /api/interactions/check if the answer mentions any drug+supplement pair.
6. Render the response with linkified PMIDs/DOIs.

Example: ground a magnesium-sleep query

# Pull tier-A and tier-B evidence, sorted by quality score
curl -H "Authorization: Bearer $PRIOLIX_KEY" \
  "https://priolix.com/api/sources?vertical=magnesium-sleep&min_score=60&page_size=8"

# Returned fields per record:
# {
#   "pmid": "30923852",
#   "doi": "10.1093/sleep/zsy191",
#   "title": "Effect of magnesium supplementation on sleep quality ...",
#   "quality_tier": "A",
#   "quality_score": 82,
#   "evidence_level": "meta_analysis",
#   "year": 2019,
#   "abstract_or_summary": "..."
# }

Why not build it yourself?