Navigation: ML/GenAI • Robotics/Control • Software Engineering • Competitions • Projects
I thrive at the intersection of engineering domains, building complete, intelligent systems from the ground up.
Machine Learning
|
Robotics & Control Sys
|
Software Engineering
|
I build and deploy high-performance models that solve real-world problems. My approach is validated by consistent top-tier performance in global data science competitions, where I've engineered end-to-end solutions for classification, regression, and forecasting challenges. This involves the full data lifecycle: acquiring and processing data from diverse diverse sources via APIs, extensive feature engineering, and implementing sophisticated models. I leverage everything from gradient boosting machines and neural networks to advanced techniques like ensembling and multi-stage training to tackle complex problems in NLP, Computer Vision, and Generative AI.
- Battle-Tested Performance: Consistently ranked in the top 10% across multiple international data science challenges on Zindi, Solafune, and ThinkOnward.
- End-to-End MLOps: Experienced in the full ML lifecycle, from data processing to building production-grade inference pipelines using AWS SageMaker, NVIDIA Triton Inference Server, ONNX/TensorRT, and CI/CD with GitHub Actions.
- From Paper to Code: Adept at reading scientific papers on novel deep learning methods and implementing them from scratch in PyTorch.
Featured Competition Methodologies
- Clinical NLP (PATH / Zindi) — Top 6%: Fine-tuned a
google/flan-t5-basemodel for this Seq2Seq task. I engineered a structured prompt that incorporated clinician context (experience, location) with the raw medical scenario to improve response quality. The final inference pipeline utilized beam search, length penalty, and n-gram blocking to generate coherent, high-quality clinical responses, evaluated with the ROUGE metric. - Generative AI for Geophysics (ThinkOnward) — Top 7%: Developed a hybrid solution by ensembling a fine-tuned GeoFoundation model with a custom ViT-U-Net architecture. This dual-model approach effectively captured both broad geological patterns and fine-grained seismic details, leading to a high-accuracy model for geophysical velocity inversion.
- Geospatial CV (Solafune) — Top 8%: Deployed a YOLOv8 model with a multi-stage training strategy on different image sizes, using Albumentations for robust data augmentation. Critically, to refine the dataset, I built a custom annotation tool, creating a human-in-the-loop pipeline for iterative data improvement and retraining, which significantly boosted model performance.
- Geospatial Time-Series (Amini | NASA / Zindi) — Top 14%: Engineered a complex data acquisition pipeline using APIs (Harvard Dataverse, Zenodo) and Google Earth Engine to extract multispectral features (RGB, NIR, SWIR) for given geolocations. Utilized NASA's PRESTO foundation model to generate powerful time-series embeddings, which then fed a LightGBM ensemble model with isotonic calibration for highly accurate crop classification. find Notebook and Methodology Paper
- Soil Nutrients (Zindi) — Top 8%: Implemented a robust LightGBM model to tackle the multi-target regression task, successfully predicting 11 distinct soil nutrient levels simultaneously from environmental and spectral data.
My background as an Electrical Engineer allow me to bridge the digital and physical worlds by designing, building, and programming intelligent electromechanical systems from concept to completion. I don't just connect components; I engineer the software that makes them work in concert.
- Precision Motion Control: Mastered high-precision stepper motor control using advanced Trinamic drivers like the TMC2130. I develop firmware that leverages SPI to configure sophisticated features like microstepping, sensorless feedback (StallGuard™), and silent operation (StealthChop™) for NEMA-class motors.
- Industrial Automation & Control: Proficient in programming Siemens PLCs using TIA Portal and developing Human-Machine Interfaces (HMIs) with SCADA systems. Experienced in designing and commissioning automated sequences for industrial machinery and drive systems via platforms like DriveLab.
- Firmware & Sensor Integration: Authored custom firmware drivers in C/C++ for AVR and ARM microcontrollers to establish direct IC-level communication. I write the software that interfaces with a diverse range of sensors—including IMUs, temperature sensors, and magnetic encoders—and orchestrates complex, event-driven actuator sequences.
- System Architecture & Edge AI: Architected complete robotic systems by integrating high-level logic (Python, ROS) with low-level hardware control. Deployed OpenCV on Raspberry Pi for real-time vision tasks, creating a closed-loop system where sensor data directly informs control logic.
- System Design & Simulation: Proficient in creating complex 3D models and assemblies for industrial and robotic systems using AutoCAD, SolidWorks, and Inventor. I validate system dynamics and control logic through advanced simulation in MATLAB/Simulink before taking designs from schematic to reality, laying out custom multi-layer PCBs in Altium Designer for robust and reliable control boards.
Technical Project Highlights
- High-Precision Industrial Manipulator Control System (Final Academic Project ): Designed, built, and validated a control system for industrial manipulators. Developed C++ firmware for an ATmega32U4 to command a NEMA 17 motor via a TMC2130 driver. The system allows real-time SPI control of torque, speed, and microstepping, while collecting diagnostic data for sensorless feedback using StallGuard2. Authored a comprehensive technical article detailing the architecture, validation, and performance. See Here
- Autonomous Pick-and-Place Crane: Engineered an end-to-end autonomous transport system. The core logic involved real-time OpenCV object tracking on a Raspberry Pi, which commanded a 3-axis crane, conveyor belt, and electromagnetic actuator using optimized Python and GPIO sequencing.
- Custom Flight Controller Drone: Built and programmed a quadcopter drone from scratch. Developed custom flight control firmware in C++/Arduino, implementing PID loops for real-time stabilization based on data fused from an IMU (gyroscope & accelerometer), and managed motor speeds via ESCs based on RC controller inputs.
- ROS-Enabled Reconnaissance Rover: Revitalized a 6-wheel mobile robot by deploying a ROS stack on a Raspberry Pi for real-time telemetry and remote operation. Developed C++/Arduino firmware to interpret RC commands and precisely control multiple IBT-2 motor drivers in a differential drive configuration.
I develop robust, scalable, and efficient software solutions, with a deep expertise in the Java ecosystem for creating full-featured desktop and backend applications.
- Backend & Desktop Development: I build robust REST services using Spring Boot, FastAPI, and Flask, and craft rich desktop applications with JavaFX and Python (Qt5). I adhere to clean architecture principles (MVC), manage dependencies with Maven, and ensure quality with comprehensive testing.
- Automated Data Engineering: I develop automated data pipelines using Selenium and Scrapy for web data extraction and Pandas for transformation, designing for reliability with integrated job scheduling, logging, and monitoring.
- High-Performance Python with C++: For speed-critical applications, I package high-performance C++ code into Python wheels. By leveraging the CPython C-API and CMake, I create optimized modules for demanding numerical computing and ML tasks.
Featured Software Projects
- H2O Provider — Water Management Desktop App: Engineered a full-featured application to digitize and centralize reports for water distribution stations, featuring a dashboard, automated reporting, and user access controls.
- Tech: JavaFX, MVC Architecture, Multithreading, SQLite, iTextPDF for reporting.
- G-Chimie — Networked Board Game: Co-developed a turn-based board game with a full client-server architecture for multiplayer gameplay over a network.
- Tech: JavaFX, Java Sockets (Networking), MVC, Multithreading.
- Transky — Real-Time Translation Widget: Built a standalone desktop widget for real-time speech transcription and translation by integrating with the Google Translate API.
- Tech: JavaFX, REST API Integration, Asynchronous Programming.
- Master Brain — Efficient Learning Software: Developed a custom software application to aid in language learning and text memorization, designed to overcome personal learning challenges.
- Tech: JavaFX, SQLite, MVC, Data Binding.
- Geodes — Geometric Description Software: Created a software tool to facilitate the creation of 2D geometric descriptive sketches from 3D projections, aimed at architects and engineers.
- Tech: JavaFX, Custom Geometric Algorithms, MVC.
- Short Activist Predictor — Python Package: Collaborated on publishing a Python module for financial text analysis, managing the full packaging and deployment lifecycle.
- Tech: Python, PyScaffold for packaging, Twine for PyPI deployment, GitHub Actions for CI/CD.
| Competition Title | Platform & Host | Domain | Rank | Placement |
|---|---|---|---|---|
| LLM for Clinical Decision Support | Zindi / PATH | Clinical NLP / LLMs | 22 / 387 | 🥇 Top 6% |
| Generative AI for Geophysics | ThinkOnward | Geophysics / Diffusion Models | 27 / 386 | 🥇 Top 7% |
| Vacant Lot Detection | Solafune | GeoAI / CV Segmentation | 32 / 447 | 🥇 Top 8% |
| Amini Soil Prediction Challenge | Zindi / Amini | Tabular / Multi-target Regression | 26 / 308 | 🥇 Top 8% |
| Amini GeoFM Decoding the Field Challenge | Zindi / Amini & NASA | GeoAI / Time-Series | 18 / 137 | 🥈 Top 14% |
-
Hypersphere-based Fuzzy SVM Classifier (Independent Research) — Implementing a novel classification algorithm to effectively handle high class imbalance and noise.
- Tech: C/C++, Python (CPython API | Pytorch), CMake, Scikit-learn (Integration), Fuzzy Logic.
- Links: Hyperion Fuszzy (In Progress)
-
GeoFM Crop Classification — The solution for the Amini/NASA competition, involving a complex data pipeline and foundation model.
- Tech: NASA PRESTO, PyTorch, Google Earth Engine, LightGBM, GeoPandas.
- Links: Notebook • Methodology Paper
-
Exploratory Analysis of Startup Failures (Downloads 2700+) — An EDA project analyzing a dataset of failed startups to identify key contributing factors through data visualization.
- Tech: Python, Pandas, Kaggle Notebooks, Data Visualization.
- Links: Notebook• Startup Failures Dataset
-
Short Activist Predictor — A published Python module and Hugging Face model for financial text classification.
- Tech: BERTopic, scikit-learn, PyScaffold, GitHub Actions (CI/CD), PyPI.
- Links: PyPI Package • Hugging Face Model • GitHub Repo
-
High-Precision Industrial Manipulator — My final academic project, a complete control system for industrial manipulators.
- Tech: C++, Python (Qt5), ATmega32U4 Firmware, TMC2130 Driver, Nema, SPI, Control Theory, Inventor.
- Links: Technical Article • GitHub Repo
Other projects
-
Design Trend Analysis via Notion API — An end-to-end data project to extract data from Notion, analyze market trends for package design, and visualize insights.
- Tech: Python, Notion API, Pandas, Data Analysis, Matplotlib/Seaborn.
- Links: Kaggle Notebook
-
Custom Flight Controller Drone — A quadcopter built from scratch with custom flight control firmware.
- Tech: C++, Arduino, IMU Sensor Fusion, PID Control Loops, Real-Time Programming.
-
ROS-Enabled Reconnaissance Rover — A revitalized 6-wheel mobile robot with a full ROS stack for telemetry and remote operation.
- Tech: ROS, Python, C++, Arduino, Motor Drivers (IBT-2).
-
Transky — A real-time speech transcription and translation desktop widget.
- Tech: JavaFX, REST API Integration, Asynchronous Programming.
- Links: GitHub Repo
-
G-Chimie — A networked multiplayer board game with a full client-server architecture.
- Tech: JavaFX, Java Sockets (Networking), MVC, Multithreading.
- Links: GitHub Repo
-
H2O Provider — A complete water management desktop application with a dashboard, automated reporting, and user access controls.
- Tech: JavaFX, MVC Architecture, Multithreading, SQLite, iTextPDF.
- Links: GitHub Repo
- NVIDIA Certified Professional - AI Operations (NCP-AIO) (2025)
- Publication: "High-Precision Control System for Industrial Manipulators" (2025)
- Publication: "GeoFM for Crop Classification - Methodology Paper" (2025)
|
|
|
|
- LinkedIn: linkedin.com/in/daglox-kankwanda
- Website: danglchris.com
Languages: 🇫🇷 French (Native) • 🇨🇩 Lingala (Native) • 🇺🇸 English • 🇷🇺 Russian