SCAN IN.
REPORT OUT.
Security & Data Handling
What we do, what we don't, and what infrastructure providers cover. We do not claim certifications we haven't earned. Infrastructure attestations belong to Groq, Google Cloud, and Razorpay — not to OmniBioFex. Where a control is designed-for but not independently audited, we say so.
🔐 EncryptionTLS 1.2+
All traffic encrypted in transit; 1.3 negotiated wherever the client supports it.
AES-256 AT REST
Firestore + Cloud Storage encrypted with AES-256 by Google default.
GOOGLE OAUTH 2.0
Identity via Google. Zero passwords stored on our servers.
SHORT-LIVED JWT
Session tokens issued and auto-rotated by Firebase Auth.
DEFAULT-DENY RULES
Firestore blocks all reads/writes unless explicitly permitted by rule.
PER-USER SCOPING
Users may read only their own reports and wallet data.
UPLOADS ARE STORED
Uploaded images are stored in your private Cloud Storage folder so they can appear in your report history. You can request deletion at any time.
NO TRAINING
Your uploads are never used to train or fine-tune any model.
30-DAY DELETION
Account and history deletion completed within 30 days of a verified request.
NO PHI REQUIRED
Do not upload identifiable PHI. Use de-identified images unless you have independent legal authority.
HIGH-THROUGHPUT INFERENCE
Qwen 3.8 27B served on Groq's LPU stack; model card states up to ~450+ tps.
GROQ — SOC 2 TYPE II
Attestation held by Groq, not OmniBioFex.
GROQ DATA POLICY
See groq.com/privacy for their current inference data-retention policy.
PCI-DSS L1
Held by Razorpay, our payment processor.
ISO 27001:2022
Held by Razorpay.
RBI-AUTHORISED PA
Razorpay holds a Reserve Bank of India Payment Aggregator license.
SOC 1 / 2 TYPE II / 3
Held by Google Cloud.
ISO 27001 / 27017 / 27018
Held by Google Cloud.
NIST AI RMF
Design aligned. No independent audit.
EU AI ACT — ART. 4
Literacy obligations considered in product design. No independent audit.
WHO AI ETHICS
Design aligned. No independent audit.
What You Get
Every scan produces a formal clinical imaging report. Structured like a radiologist's report, exportable to a multi-page PDF.
EXAM TITLE + VIEW
Auto-detected from the image — e.g. "CHEST X-RAY · PA View", "CT BRAIN · Axial", "MRI KNEE · Sagittal".
STRUCTURED FINDINGS
6–12 discrete clinical sentences covering anatomy, margins, densities, and any abnormalities.
IMPRESSION
One-sentence diagnostic conclusion. "No abnormality detected" when normal, or the leading diagnosis otherwise.
CLINICAL ADVICE
One-line recommendation — often "clinical correlation recommended" or a specific next step.
DIFFERENTIAL DIAGNOSIS
3–6 conditions ranked most-to-least likely, each with a short rationale.
MANAGEMENT
3–6 specific next steps — additional imaging, labs, referrals, or conservative management.
CLINICAL PDF EXPORT
Multi-page PDF with letterhead, patient details, your scan image, and all report sections.
AUTO-SAVED HISTORY
Every report is stored under your account with patient name, case ID, and timestamp — retrievable anytime.
How It Works
Sign in with Google. Attach a scan. Enter the patient's name, age, and gender. That's it — the pipeline does the rest in 15–30 seconds.
SIGN IN
One-click Google OAuth. No passwords, no email verification, no credit card. Account is live the moment you consent.
ATTACH SCAN
Drop a chest X-ray, CT, MRI, or histopathology slide. Full image is used — no cropping. Multi-image uploads for longitudinal comparison.
ENTER PATIENT DETAILS
Name, age, and gender. Required for the report letterhead. Stored alongside the report in your private history.
AI PIPELINE RUNS
Four sequential steps: vision extraction → impression → differential + management → final report assembly.
REPORT + PDF
The report appears in chat. Click EXPORT PDF to download a clinical-grade document with your patient details, the scan image, and all report sections.
Capabilities
One model. One report format. All output is decision-support material for qualified professionals — never a diagnosis.
01 · CHEST X-RAY
Descriptive observations of lung fields, cardiac silhouette, mediastinum, pleural spaces, and osseous structures. Findings must be verified by a radiologist.
02 · CT — SINGLE FRAME
Descriptive observations on single CT slices. The model does not reconstruct 3D volumes.
03 · MRI — SINGLE FRAME
Descriptive observations on single MRI frames. Not a substitute for a radiologist's read.
04 · MUSCULOSKELETAL
Descriptive observations on radiographs. Not a fracture-detection product; findings require professional review.
05 · TISSUE IMAGING
Descriptive observations on tissue images. Not a pathology AI; output must be reviewed by a pathologist.
06 · SKIN IMAGING
Descriptive observations on skin lesion images. Output must be reviewed by a dermatologist before any clinical use.
Specification Sheet
Qwen 3.8 27B is a 27-billion-parameter multimodal model from Alibaba's Qwen series, serving both image and text inputs with native JSON mode and strong instruction following.
| Attribute | Specification |
|---|---|
| Architecture | Hybrid Gated DeltaNet + Gated Attention |
| Input modalities | Text, images (max 3 per request, 2048 tokens/image) |
| Output modality | Text only (JSON-native) |
| Context length | 131K tokens |
| Reasoning modes | Thinking + Instruct (switchable per request) |
| Tool use | Function calling + tool orchestration |
| Multilingual | Strong multilingual support (EN, ZH, and more) |
| GPQA Diamond | 89.2% |
| LiveCodeBench v6 | 90.3% |
| IFBench (instruction following) | 79.5% |
Ready to start?
Start with the free tier, or top up a wallet and skip the cooldown. No card stored. No auto-renewal.