Marco Mongi

License Plate Detector

Real-time pipeline for license-plate detection, multi-object tracking and OCR, with an interactive UI and a fully reproducible, production-oriented setup.

Try online

AutoCBC

Automated blood cell counting system built with state-of-the-art computer vision. Combines YOLO object detection and SAM2 instance segmentation to accurately detect, classify, and segment RBCs, WBCs, and platelets from microscopy images. Provides real-time interactive overlays and comprehensive quantitative analysis via a Streamlit web interface.

Try online →

Blood Cells Count Aid

Automated microscopy assistant that detects and counts red blood cells, white blood cells, and platelets in stained blood-smear images, overlaying colour-coded markers in a PyQt5 desktop UI-achieving ≈ 98 % overall counting accuracy versus expert annotations on selected test images.

CNN Digits Detector

Interactive handwritten digit recognition system featuring a Convolutional Neural Network trained on the MNIST dataset with image augmentations. Provides real-time inference through an interactive sketchpad interface - achieving 99.49% accuracy on test data.

Try online

OpenGPT Chat

Self-hosted, fully offline AI chat platform that lets you interact with local LLMs, tools, and private documents. Combines RAG, agent workflows, voice input, and a conversational UI into a single Docker-based open-source stack with full data privacy.

• Local LLM Inference (Ollama)
• Retrieval-Augmented Generation (RAG)
• Agentic AI Workflow (n8n)
• Vector Database (Qdrant)
• PostgreSQL
• Docker Compose
• Conversational UI (Open WebUI)

Resume Refiner Crew

AI-powered resume optimization using multi-agent intelligence. Seven specialized CrewAI agents collaborate to transform your resume into a job-specific, ATS-optimized document with Harvard-style PDF formatting-analyzing job fit, incorporating keywords, fact-checking content, and controlling length while maintaining truthfulness.

• Multi-Agent Systems (CrewAI)
• LLM API interaction (OpenAI)
• Resume Parsing & Optimization
• LaTeX PDF Generation
• Streamlit UI
• Docker

Try online →

Interactive Adventure Generator

Agentic storytelling app where an LLM narrates a branching adventure and adapts the plot to the player's choices. Supports bilingual (EN, SPA) voice & text interaction, streaming responses, and configurable model providers (API Key or local).

• LLM API interaction
• Prompting
• Agentic flow (LangChain)
• TTS (Piper)
• STT (Whisper)
• Gradio UI

Try online

Whisper Studio

Graphical interface for OpenAI's Whisper speech-to-text engine. Transcribe and translate audio/video files locally with support for 99 languages, 6 model sizes, microphone recording, and multiple output formats (txt, srt, vtt, json)-all processing happens on-device with no external APIs.

• Speech-to-Text (Whisper)
• Streamlit UI
• Docker

Health Anomaly Detector

End-to-end wearable-style monitoring stack that simulates vital signs, detects user-specific anomalies with unsupervised learning, and streams metrics and alarms into Grafana dashboards via Telegraf + InfluxDB. Includes a FastAPI service, Streamlit controls, Docker Compose, and AWS deployment.

Simulation Monitoring Board

Lunar Lander RL Controller

Interactive application combining deep reinforcement learning with human control in Gymnasium's LunarLander-v3 environment. Features a pre-trained Deep Q-Network agent that learned optimal landing strategies through 1M training timesteps, alongside manual control mode with seamless AI-human switching via terminal interface.

• Reinforcement Learning (DQN)
• Gymnasium
• Stable-Baselines3
• Terminal UI
• Docker

Survival Probability Simulator

Interactive ML app that predicts survival probability from demographic and socioeconomic features using a Random Forest model, with real-time Streamlit predictions, comprehensive EDA with interactive visualizations, and a Dockerized data pipeline.

• Machine Learning (Classification)
• Streamlit UI
• EDA & ETL
• Docker

Try online

Household Energy Forecasting

End-to-end data-science project forecasting household electricity consumption, covering EDA, preprocessing, feature engineering, time-series modeling, comparison, and evaluation, with results documented in a scientific report and presentation (Spanish).

• EDA & ETL
• Feature engineering
• Time-series models: FFNN, Prophet, RF
• Fourier seasonality
• Gradio notebooks
• LaTeX reporting
• Jupyter / Colab

Market-Sentiment Predictor

End-to-end NLP and time-series pipeline that scrapes finance tweets, extracts daily sentiment (BERT-based), fuses it with market prices, and trains classification and regression models for next-day moves - revealing near-random predictive power.

WeatherFlow

Docker-based Apache Airflow pipeline that orchestrates real-time weather data collection and processing. Implements a complete ETL workflow with API monitoring sensors, data extraction from OpenWeatherMap, transformation into normalized database structures, and bulk loading into PostgreSQL-running automatically every 10 minutes with failure handling and duplicate prevention.

• Apache Airflow
• ETL & Data Engineering
• Docker & Docker Compose
• SQL

OEE Analytics

Industrial edge-AI system for scalable production-line monitoring, using multiple cameras positioned along a conveyor to detect, track, and count items in real time on embedded hardware. Vision-derived counts are combined with additional operational inputs to compute real-time and historical OEE metrics, enabling continuous monitoring of throughput, availability, and performance across the line.

SABIAMar L0 Processor

Enhanced the SABIA-Mar Level-0 processor by decoding raw satellite data from transport layers down to pixel-level images and integrating telemetry metadata, forming the foundation of the satellite image processing pipeline.

• Python
• Automated testing
• Binary-frame decoding
• Data-pipeline design

Satellite Crop Detector

Python-based system for crop-type classification from multi-spectral satellite time series, using classical computer-vision preprocessing and engineered temporal features with a Random Forest model, achieving a 93% F1 score.

Poultry Environment Regulator

Automated control system for chick brooders, using a microcontroller to manage heating, ventilation, and lighting schedules. Based on Arduino UNO with DHT11 sensing, RTC time-keeping, and relay-controlled loads. Settings are adjusted via an LCD-encoder menu, with safe defaults restored after power loss.

Tablet-Joystick Dock

Custom 3D-printed add-on that clamps a joystick to a tablet while rerouting power, audio and volume controls through internal USB / 3.5 mm extensions and a dual-button circuit-restoring full charging, headphone and volume functionality during gameplay.

• 3D design
• 3D printing
• Electronics prototyping