Ali Mobini.

A movie recommendation app I built and run as sole developer, live and free at cinreco.com: describe your mood in plain English, swipe to teach it your taste, and get shared picks for your whole group.

Key Highlights:

• FastAPI, PostgreSQL/pgvector, Redis and Celery backend with 39 REST endpoints and 56 automated tests.
• 5-stage hybrid recommender: vector retrieval, knowledge-graph boost from TMDB cast and crew, temporal re-ranking, cross-encoder re-scoring and MMR diversification.
• Group recommendations ranked by a least-misery consensus score, with a plain-language reason for every pick.
• Hardened auth (refresh-token rotation with reuse detection, Argon2, email verification, lockout, Redis rate limiting) and hourly TMDB catalog ingestion.

The CinReco Flutter app, ported from mobile to the web and shipped with a full delivery pipeline. Live at app.cinreco.com, with a public demo at demo1.mobini.nl.

Key Highlights:

• Multi-stage Docker build to a 143 MB non-root nginx image; runtime config injection lets one image serve both demo and production.
• GitHub Actions CI (format, analyze, test, image build, smoke test), GHCR releases and signed Android APK/AAB builds.
• Pull-based deployment: a systemd timer polls GHCR every 5 minutes and applies Terraform only when the image digest changes, so no deploy credentials leave the server.
• Terraform-managed Traefik, Prometheus and Grafana (dashboard as code); first hosted on my homelab over Tailscale, now on a Hetzner VPS.

An agent skill and Claude Code plugin (MIT) that turns a project, website or idea into ready-to-post social-media videos with music, captions and post copy for each platform. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI and opencode.

Key Highlights:

• Agent workflow that writes a brief, pitches and scores video ideas across 20 marketing angles, then storyboards, builds and renders the chosen one.
• Playwright-to-FFmpeg renderer that outputs any aspect ratio from one HTML composition.
• Procedural soundtrack with voice ducking and -14 LUFS loudness normalization, plus local Kokoro-82M text-to-speech with word-by-word captions.

Graduation internship and bachelor’s thesis at AMI Research Group (Feb - Jul 2026): an EU-compliant, hardware-backed security system on NXP FRDM-RW612 (ARM Cortex-M33 + TrustZone + EdgeLock crypto accelerator) for Digital Product Passport sensor data, per NIS2 and ESPR.

Key Highlights:

• Multi-layer cryptography stack: ECDSA P-256 hardware signing, SHA-256 hashing, ECDH + AES-256-CBC encryption, TLS 1.3 via mbedTLS/PSA Crypto on FreeRTOS.
• Python Flask backend for device registration, ECDSA signature verification, replay attack detection (nonce caching), and hash-chain integrity checks.
• Immutable audit trails via IOTA Tangle anchoring and Decentralized Identifiers (DIDs) for full device lifecycle management.

GNSS-based geofence safety system for excavators with RTK correction and probabilistic decision-making.

Key Features:

• ~20cm positional accuracy with RTK
• 30% reduction in false alarms
• 100% reliable startup and >60% less downtime with a modular Linux service architecture
• Real-time TCP/WebSocket streaming

Probabilistic model improving decision reliability in GNSS-based geofence safety systems.

• Kalman filtering for GNSS smoothing and uncertainty modeling
• Bayesian-inspired likelihood and confidence estimation (PDF/CDF)
• ~30% reduction in false alarms; ~20cm accuracy with RTK

Internship at AMI Research Group (Sep 2025 - Jan 2026): an end-to-end MLOps pipeline for environmental e-nose data modeling and deployment.

• Python, Docker, Airflow, MLflow, Grafana, InfluxDB
• Ingestion, preprocessing, training (XGBoost/Chronos), CI/CD deployment
• Achieved RMSE 0.15; 40% faster model deployment time
• n8n chatbot that serves the model’s predictions

Embedded vision project at Saxion (academic year 2024-2025): a logo detector and a face-tracking robot.

• MobileNetV2 transfer learning (TensorFlow/Keras) with a 1,000-image synthetic dataset and FP16 mixed precision: 91.2% validation accuracy
• Face recognition (98.7% under optimal conditions) driving servo eye-tracking on a Raspberry Pi
• Flask web dashboard for logo detection and a desktop GUI for the robot

Team project at Saxion (Sep 2024 - Jan 2025): a precision robot with physics-based simulation and closed-loop control. Ranked 1st in our community.

• 0.025 mm (25 µm) placement accuracy against a 50 µm goal, with ANSYS-assisted modeling and trajectory planning
• My part: PyQt6 operator screens with live plots, and the ROS2 nodes linking the UI to the motor controller
• Worked with CANopen and fixed segmentation faults in the C++ motor code

University project: the face-recognition service for a door-lock system.

• Flask REST API that recognizes faces (with their positions) and enrolls new people from a photo
• OpenCV webcam CLI and a PyQt6 desktop client
• The recognition module was later reused in my eye-tracking robot

Built with AI coding agents: a minimal URL shortener designed to run on a single small VPS.

• FastAPI, PostgreSQL and Redis, with Redis-cached redirects
• SSRF guard, URLhaus malware blocklist, per-IP rate limiting and bcrypt-hashed deletion tokens
• 7 languages with GeoIP locale detection; 38 tests and GitHub Actions CI/CD to a VPS

Built with an AI coding agent: a dependency-free Bash toolkit that audits and hardens Linux servers (private repository).

• 219 checks across 18 modules: SSH, firewall, PAM, kernel, auditd, AIDE and more
• SSH lockout protection, dry runs, idempotent runs and full rollback
• CI hardens and rolls back real Ubuntu 22.04/24.04 runners

Led 6-member team designing an autonomous RC car with microcontrollers, ToF sensors and PID control.

Achievements:

• 98% path accuracy
• Obstacle avoidance & line tracking
• Adaptive hill-climbing capabilities

Co-developed an IoT wearable using ESP32 for real-time health monitoring with fall detection and biometric data collection.

Capabilities:

• 90% fall detection accuracy
• Heart rate & temperature monitoring
• OTA firmware updates

Sensor nodes connected to an interactive dashboard for real-time environmental analytics.

• Python backend with MQTT messaging
• Azure IoT Hub integration and live visualization

Full-stack e-commerce platform with JWT authentication, dynamic cart management, and role-based admin controls.

Features:

• .NET Core backend with SQL database
• Stress-tested with 500+ simulated transactions
• MVVM design pattern and dependency injection

VHDL ALU with basic instructions and a UART with VGA generator for real-time text rendering.

• Designed and synthesized ALU instructions in VHDL
• Implemented UART protocol and 1080p/60Hz VGA timing
• Timing analysis and debugging with Quartus toolchain

Soft-core processor integration with custom FPGA peripherals and embedded C validation.

• NIOS II system integration and peripheral mapping
• Embedded C programs for hardware validation and I/O testing

Custom Linux character device driver for hardware–software communication.

• I2C/SPI user-kernel interface
• Concurrency and memory errors resolved with GDB debugging and systematic testing