Product
A retrieval-first system for local guidance.
GuidelinesIQ is designed for local protocols, not generic web search. It treats figures, tables, and flowcharts as first-class evidence and keeps generation constrained to retrieved context.
How it works
Ingest
Upload PDFs. GuidelinesIQ parses text, tables, figures, and flowcharts, including scanned pages when needed.
Chunk and embed
Semantic chunking preserves structure while figures are extracted as retrievable units with page metadata.
Retrieve and rerank
Dense retrieval narrows candidates, reranking sharpens relevance, and the system can reformulate queries when evidence is thin.
Generate and cite
A frontier-class LLM generates a source-grounded summary constrained to retrieved evidence with citations attached to claims.
What’s under the hood
Figure-aware retrieval
Agentic multi-step query reformulation
Citation grounding with excerpt spans
Hybrid dense and sparse search
Role-tuned system prompts
Continuous benchmark harness
Deployment and operating posture
SaaS
Fastest path to evaluation for non-PHI local guideline access. Dedicated tenant architecture, managed inference, and exportable audit logs.
On-prem or VPC
For institutions that require full data sovereignty. GuidelinesIQ runs on your infrastructure, your keys, and your audit pipeline while preserving the same non-PHI posture.
Built for guideline access, not chatbot theater.
The product is intentionally framed as informational access acceleration. That makes the evidence trail inspectable and keeps responsibility with the treating clinician.