Skip to main content
Beta Test

Find Your Rare Diagnosis in Minutes.

Our AI analyzes your symptoms against thousands of conditions, focusing on rare and complex diagnoses that might be overlooked by general practitioners.

Start My Health Analysis

Free during early access

DIAGNOSTIC ANALYSIS

Whether you're searching for a medical second opinion, exploring an AI symptom checker for rare diseases, or looking for help with a complex diagnosis, SecondLook provides the analytical depth that standard tools lack. Our platform is designed for patients navigating a diagnostic odyssey who need more than generic health advice.

Important Medical Notice

This analysis is AI-generated and is for educational purposes only. It does not replace professional medical advice, diagnosis, or treatment. Always consult with qualified healthcare providers for medical decisions, especially before acting on any AI-suggested diagnosis or test.

How SecondLook Works

A multi-stage diagnostic pipeline that turns your story into a ranked differential and concrete next steps.

Step 1
Tell us your medical story and upload relevant history and data
Step 2
SecondLook extracts clinical concepts
Step 3
Map symptoms and concepts to candidates in our 9k+ rare disease knowledge base
Step 4
Activate 5 most relevant AI specialist agents in parallel to debate and select most likely diagnoses from profile
Step 5
Synthesize and rank a top-10 differential diagnosis list
Step 6
Refine diagnoses with 3–5 targeted patient questions
Step 7
Finalize the top-10 differential and probabilities
Step 8
Recommend tests to rule diagnoses in or out
Step 9
Deliver the final report
Benchmark Results

SecondLook beats current benchmarks in rare disease diagnosis

Tested against the same Phenopacket2Prompt benchmark used to evaluate o1-preview and GPT-4o in the peer-reviewed literature — graded the same way, in Mondo ontology space, so the numbers are directly comparable.

head-to-head
+20.8%
More Top-3 accuracy than OpenAI o3 single-shot on identical cases — same patient, same grader.
head-to-head
+15.5%
More Top-3 accuracy than Claude Opus 4.7 single-shot on identical cases — same patient, same grader.
vs. prior LLM SOTA
+55.5%
More Top-1 accuracy than o1-preview from the published Phenopacket2Prompt evaluation.

About the Phenopacket2Prompt benchmark

Phenopacket2Prompt is a public dataset of 9,587 published clinical vignettes, each derived from a peer-reviewed case report and paired with a verified ground-truth diagnosis (typically an OMIM identifier). Because every case maps to a real published patient, it is widely used as the rare-disease benchmark for diagnostic AI evaluation.

The Claude Opus 4.7 and OpenAI o3 numbers were generated by us on the same case sample as SecondLook, using each model in a single-shot diagnostic prompt so the comparison is head-to-head — apples-to-apples LLM evaluation throughout.

References & methodology
  1. Robinson PN et al. (2026). Evaluation of LLMs on rare-disease diagnosis using the Phenopacket2Prompt benchmark. European Journal of Human Genetics. o1-preview and GPT-4o Top-N rates reported on n=5,213 cases.
  2. Head-to-head Claude Opus 4.7 and OpenAI o3 numbers measured by SecondLook on the same random sample as our pipeline (n=30), using each model in a single-shot diagnostic prompt against the same vignettes.
  3. Phenopacket2Prompt dataset: doi:10.5281/zenodo.15065293.
  4. Grading is paper-faithful: each prediction is grounded to a Mondo class, scored 1.0 for an exact OMIM/skos:exactMatch hit and 0.5 for an IS_A ancestor of the gold; Top-N counts a case correct when any of the top N has score > 0 (Robinson et al. methodology).

Health Resources & Rare Disease Guides

Explore our guides on rare disease diagnosis, navigating complex medical cases, and making the most of AI symptom checkers on your health journey.

Ready to explore your health?

Get AI-powered diagnostic insights for complex and rare conditions — free during early access.

Start My Health Analysis