About

What we do

No B.S. Med retrieves facts from clinical studies for an AI agent to reason over in an auditable manner by humans.

The problem No B.S. Med is attacking

The problem we are attacking is that it's hard to audit the reasoning of ChatGPT's health answers.

What we do differently than ChatGPT and Google Search

ChatGPT gives you reasoning with some citations. Google Search gives you a massive list of all relevant web page citations. No B.S. Med returns medical evidence as an Audit Bundle.

What is an Audit Bundle?

Each AI health claim reasoned from our data can be linked to an Audit Bundle: the evidence subgraph, query trace, and supporting study records behind the answer.

Audit Bundle	What it is	Example
Claim	A single sentence in an answer — a specific assertion about an intervention, condition, or outcome.	“Intensive blood pressure control reduced stroke incidence in adults at elevated cardiovascular risk.”
Evidence Subgraph	The minimal slice of the knowledge graph that supports the claim — interventions, conditions, outcomes, mechanisms, and the findings that connect them.
Supporting Study Records	The underlying manuscripts (PMC IDs, titles, links) where each finding was extracted.	PMC4689591 — SPRINT Research Group, NEJM 2015. Finding positive Intensive BP control (target <120 mmHg) reduced stroke incidence vs. standard control (<140 mmHg). Intervention Intensive BP control Outcome Stroke Effect size HR 0.89 Sample 9,361 Design RCT PMC4689591 SPRINT — A Randomized Trial of Intensive versus Standard Blood-Pressure Control
Query Trace	The Cypher query and intermediate derivation steps the system ran to assemble the evidence subgraph. Helpful because your AI's claim might be based on false negatives — the user should be able to audit what the net threw out.	MATCH (i:Intervention)-[:tested]->(f:Finding)-[:on_outcome]->(o:Outcome) WHERE i.name CONTAINS 'blood pressure' AND o.name CONTAINS 'stroke' RETURN f LIMIT 100 247 candidate findings matched 89 passed pertinence filter 12 cited in the final answer

Why a map view (aka. network view or knowledge graph view) of the medical evidence?

Think of Google Maps navigation. It doesn't just tell you how to get there — it shows you the surrounding map, so you can audit the suggested route and often override it with your own reasoning (“skip the highway 5 to LA, too many bad side-effects on my mood, the choice is between 101 or coast”). That hybrid of machine optimization plus human judgment only works because the underlying map is visible.

A map view shows how every single local object is completely inter-connected with every other object — its relevance is completely inter-dependent on others. In other words, a human or AI agent can explore/trace/crawl the knowledge just like they crawl the web.

How we built it

We compiled NIH (National Institutes of Health) funded manuscripts into a medical evidence knowledge graph, with standardized biomedical terminologies (LOINC, SNOMED CT, MeSH, CUIs).