IrisGuard is a high-fidelity full-stack authentication prototype developed as an advanced academic capstone project. The platform establishes a biometric verification pipeline by combining real-time face matching vectors with computer vision eye tracking overlays, interactive HUD generation, and distributed third-party notification telemetry.

Designed as a rapid-turnaround proof-of-concept, the architecture prioritizes modular decoupling, decoupled fail-safe exception boundaries, and multi-channel notification streaming.

• Technical Architecture & Highlights Computer Vision & Biometric Alignment: Architected an automated image processing pipeline using face_recognition (dlib bindings) to extract 128-dimension facial embedding vectors. Computes linear Euclidean metrics against enrolled profiles in Firestore with a custom strict distance threshold to accurately flag authorized identities versus intruders.

Dynamic HUD Processing & Extraction: Leverages OpenCV (cv2) to compute visual matrix coordinates, track eye positions, extract localized crops for audit records, and dynamically stream Base64 frame payloads between the Next.js client UI and the FastAPI backend.

Fail-Safe & Non-Blocking Workflow Design: The core verification endpoints utilize a resilient execution structure. Secondary actions—such as localized image cropping, cloud media streaming, real-time telemetry logging, and external communication links—are isolated in fail-safe boundaries to ensure identity evaluation succeeds even if external network dependencies fail.

Modular Multi-Factor Backend (FastAPI): Formulated a structured Python API utilizing explicit router segregation across segregated modules:

/iris: Handles face and eye telemetry matching.

/otp: Orchestrates multi-factor text/email challenges.

/enroll: Controls new profile onboard embedding data creation.

/admin: Administers platform configurations.

Multi-Channel Alert Telemetry: Integrates a diverse notification layer that instantly dispatches access snapshots and confidence diagnostics to a secure Telegram Bot client channel on breach events, supported by alternative delivery channels via Twilio SMS and Brevo Email bindings.

Robust Frontend Streaming (Next.js): Configured a decoupled Next.js web interface mapped to live system canvases to pull camera device feeds, format low-latency payloads, and display dynamic visual HUD overrides compiled directly by backend calculations.

• Core Subsystem Layout Plaintext ├── backend/ │ ├── main.py # FastAPI initialization & global CORS middleware │ ├── routers/ # Segmented REST API routes (iris, otp, admin, enroll) │ ├── services/ │ │ ├── iris_detector.py # Vector comparison & facial embedding computation │ │ ├── eye_overlay.py # Bounding box calculation & OpenCV HUD injection │ │ ├── storage_service.py # Cloud persistence management │ │ └── telegram_service.py# Remote event notification streaming │ └── requirements.txt # Unified Python dependencies tracking └── iris-detection/ # Next.js streaming user interface client application

• Key Engineering Practices Demonstrated Decoupled Data Flow: Biometric data remains ephemeral; images are processed in-memory as Base64 strings, and file conversions occur on-the-fly to reduce storage overhead and keep transactions efficient.

Defensive API Engineering: Implemented strict route configurations, comprehensive global CORS policies, and explicit JSON output schemas (IrisScanResponse) to ensure clean integration across platforms.

Enterprise Third-Party Integrations: Successfully unified disparate software ecosystems—ranging from cloud-native persistence layers (Firebase) to instant notification services (Telegram, Twilio, Brevo)—into a coherent data management pipeline.