Build Medical AI SaaS for Fetal Ultrasounds -- 2

Project Title: AI-Powered Fetal Ultrasound Biometry SaaS Platform
Type: Full-Stack Medical AI Web Application

We are looking for an experienced full-stack developer with medical AI expertise to build a cloud-hosted SaaS platform that analyzes fetal ultrasound images (DICOM/PNG/JPG) using deep learning, extracts biometry measurements, and generates branded medical-grade PDF reports for patients, doctors, and hospitals.
Phase-1 AI Scope (HC18 Model — Fully Supported): The current AI model is trained on the HC18 fetal head ultrasound dataset. Phase-1 directly infers ellipse parameters (Center X, Center Y, Semi-major axis, Semi-minor axis, Rotation angle θ) and derives the following measurements: Head Circumference (HC), Biparietal Diameter (BPD), Occipito-Frontal Diameter (OFD), and Head Area. From these, the platform will also compute Gestational Age (GA) from HC reference charts, Estimated Fetal Weight (EFW) using the Hadlock formula, EFW Percentile using INTERGROWTH/WHO normative tables, HC/AC ratio, FL/AC ratio, and FL/HC ratio (the last three become active once multi-parameter phases unlock).
Partially Solvable Parameters (Phase-2 — Needs Hospital Fine-Tuning): The following parameters have limited public dataset support and will require clinical fine-tuning and hospital data augmentation in Phase-2: Crown-Rump Length (CRL) via sagittal plane detection, Abdominal Circumference (AC) via FETAL_PLANES_DB and FPUS23, Femur Length (FL) via FETAL_PLANES_DB, Nuchal Translucency (NT) from sagittal plane datasets with high-precision labels, Lateral Ventricle width (LV/Atrial width) from head annotations, Transcerebellar Diameter (TCD) from brain-plane datasets, Cerebellum size from posterior-fossa crops, and Nasal Bone presence/length (NB) which is plane- and ethnically-sensitive and requires significant annotated clinical data.
Parameters Requiring Specialist Data (Phase-3+ — Not Solvable from Public Data Alone): The following parameters require Doppler waveform data, volumetric sweeps, or rich specialist annotations and cannot be delivered without hospital or R&D data collection: Amniotic Fluid Index (AFI), Single Deepest Pocket (SDP), Liquor Volume (volumetric/3D), Umbilical Artery Doppler indices (UA PI, UA S/D ratio, UA RI, UA AEDF), Middle Cerebral Artery Doppler indices (MCA PSV, MCA PI, MCA RI), Cerebroplacental Ratio (CPR), Fetal Heart Rate (FHR) from Doppler/M-mode, Humerus Length (HL), Inter-Orbital Diameter (IOD), Bi-Orbital Diameter (BOD), Placental Grade, and Placenta volumetry. These are explicitly excluded from Phase-1 and require separate datasets, dedicated AI model pipelines, and clinical validation before they can be added in future phases.
PDF Reports: The system must auto-generate three report types after AI inference — a Patient Report (HC/BPD/OFD values, GA estimate, growth percentile curve at 3rd/50th/97th, AI ellipse overlay, abnormality flags, and a mandatory AI disclaimer), a Doctor Report (original + segmentation overlay, ellipse parameter table, DICOM metadata, and doctor signature space), and a Technical Report (preprocessing pipeline logs covering DICOM to PNG/JPG conversion, resizing, normalization and windowing, inference runtime, GPU/CPU info, environment versions, and metadata checksum). All PDFs must support 8 languages (English, Hindi, Gujarati, Tamil, Telugu, Arabic, French, Spanish) via Google Translate API, render at 300 DPI, and be fully customizable per hospital with logo, color theme, watermark, and password protection options.
Platform Features: The system must include a hospital admin dashboard with role-based access control (Patient, Doctor, Technician, Hospital Admin, Super Admin), immutable audit logs, and consent logging. Security requirements include AES-256 encryption at rest, TLS 1.2+, virus scanning on uploads, and auto DICOM anonymization. The platform must integrate a ChatGPT/LLaMA AI assistant for patient-friendly report explanations and structured doctor notes (Impression, Recommendations, Observations), with a strict rule that no PHI or DICOM pixels are sent externally without hospital admin consent. Payment integration in Phase-1 covers Stripe and Razorpay/UPI only, with webhook-verified transactions and a refund dashboard. The frontend must be fully mobile-responsive using React and Tailwind, supporting camera and DICOM file uploads with offline queue handling.
Hospital Integration: Secure REST APIs must support DICOM C-STORE compatibility, FHIR REST and HL7 v2 interfaces for EMR connectivity, and batch processing for ZIP uploads of 100+ DICOM studies with background queue and status tracking. Direct PACS installation, HL7 middleware deployment inside hospital systems, and EMR customization are outside developer scope — the platform provides API endpoints only.
Infrastructure: The platform must be deployable on AWS/GCP/Azure with multi-region replication (India, US, Europe), CDN caching via CloudFront or Cloudflare, auto-scaling backend with optional GPU inference scaling, and monitoring via CloudWatch or StackDriver.
Milestones: The project follows a phased milestone delivery structure — starting with project kickoff and SRS sign-off, followed by the core AI pipeline and head biometry outputs, then multi-level PDF report generation, then the dashboard and hospital admin panel, and finally payment integration and full documentation handover. Each milestone is reviewed and accepted before moving to the next.