>Arshia Pessaran

• ├── about/
  │ └──

  Biology degree from UCF and went on to go to medical school at FIU. Withdrew to pivot into tech. Finished a second degree in data science in the span of a year and working as a software engineer ever since.

• ├── connect/
  • │ ├── github
  • │ ├── linkedin
  • │ ├── email
  • │ └── resume
• └── projects/
  • ├── Bludly
    │ └──

    A HIPAA-compliant fullstack health data platform for uploading, tracking, and analyzing blood lab results over time. Interactive D3.js visualizations chart biomarker trends with color-coded reference ranges, date filtering, and multi-lab overlay. AI analysis runs on a self-hosted Deepseek R1 model — no patient data leaves the infrastructure — generating health grades, biological age estimates, and actionable insights.
    
    Supports multiple user profiles for family management, secure time-limited doctor sharing via tokenized links, and data export. Authenticated with Google OAuth 2.0 and deployed on AWS EC2.
    
    React · Node.js · Express · PostgreSQL · D3.js · Google OAuth 2.0 · Passport.js · REST API · Docker · AWS EC2 · Deepseek R1 · Vite · Tailwind CSS
  • ├── Nautiq
    │ └──

    The officially endorsed iOS and Android companion app for the Deep6 Excursion dive computer, built as a replacement for the 1-star-rated competitor. Shipped to TestFlight and pitched to the manufacturer, securing official endorsement as their companion app with full branding and firmware collaboration rights. Reverse-engineered the BLE protocol by studying libdivecomputer C sources, sniffing a separate app's Android traffic via HCI snoop logs, and mapping the full GATT service table on physical hardware using nRF Connect and LightBlue. Discovered undocumented behavior — the device uses a single bidirectional characteristic (FFE1) rather than separate TX/RX channels, and advertises short 16-bit UUIDs that don't match the 128-bit service UUID, requiring name-based scan filtering.
    
    Implements the full TI CC2652R1 communication protocol with two firmware parser variants (V0 legacy and V1 modern), packet framing with additive-sum checksums, and fingerprint-based deduplication. Features over-the-air firmware updates via an undocumented FFE2 channel activated by a magic trigger sequence — a capability no existing app has delivered for this hardware on iOS. Firmware binaries are served from GitHub Releases with SHA-256 integrity verification and offline caching.
    
    Extracted competitor firmware binaries from Android APK assets to enable cross-version testing. Freemium model with StoreKit 2 and Google Play Billing in-app purchase. Verified against 77 dives downloaded from a physical D01-5C device. Zero third-party dependencies on both platforms.
    
    Swift 6 · SwiftUI · SwiftData · Kotlin · Jetpack Compose · Room · CoreBluetooth · Android BLE · Swift Charts · MapKit · CloudKit · StoreKit 2 · CryptoKit · BLE protocol reverse engineering · binary parsing

    
  • ├── PDF Reconstruction Pipeline
    │ └──

    A multi-stage pipeline that reconstructs scanned PDFs into clean digital documents. Preprocessing corrects page skew via projection profile analysis and Otsu thresholding, denoises with a bilateral filter to preserve edges, and binarizes for stroke analysis. Pages are classified in two passes — a fast pixel statistics pass filters obvious outliers, while a full pass combines Surya's deep learning layout model with OCR confidence metrics to label each page as normal, mixed, or outlier.
    
    A two-phase calibration system processes the first batch of pages to establish a font size and style baseline before full reconstruction. OCR is batched across pages for GPU efficiency, with support for three engines: Surya, Tesseract, and PaddleOCR.
    
    Python · PyMuPDF · ReportLab · OpenCV · Surya · PaddleOCR · Tesseract · deep learning · OCR · computer vision · image processing · pipeline architecture · customtkinter
  • └── TradingView Heatmap Bot
    └──

    A Windows background application that automates hourly capture of the TradingView S&P 500 heatmap and delivers it to a Discord channel. Uses Playwright to launch a headless Chromium browser, navigate to TradingView, configure the heatmap view, and capture a screenshot — all without user interaction. A time-aligned scheduler fires on the hour, and the captured image is posted via Discord's incoming webhook REST API.
    
    Packaged as a standalone Windows executable with PyInstaller. Runs as a persistent system tray process with pause, resume, and manual trigger controls.
    
    Python · Playwright · headless browser automation · Discord API · webhook integration · task scheduling · PyInstaller · Windows system tray