Trained multiple models to detect potholes, speedbumps, road problems, and highlight them
with a color based on severity assessment.

Tools: Keras, Mmdetection, Tensorflow, Python

Read shapes, then pack them tightly together respecting a set of rules to avoid collision such
that they are cut using fewer materials, through a shorter time span or cut from marble.
Applications are for CNC machining, 3D printing, marble cutting, art, etc.

Tools: shapely, python, numpy, pandas, ezdxf

Extrapolating 3D features from a single 2D image using a neural network to overcome the
main inaccuracies of conventional solutions.
Use 5 different skin as well as facial features to match on.

Tools: Pytorch, Python, Numpy, Pandas, Pyredner

Develop String matching algorithms using multiple methods to match 1M+ data records to
provide a coherent database for the website.
Automating data scraping into a DB, triggering the matching algorithm and indexing its
output.

Tools: Pysql, Elasticsearch, Mongo, Postgres, Gmail, Python

Solve a NP-hard route optimization problem minimizing costs and maximizing profit while
assigning each user his adequate nodes obeying each node’s constraints.
Respected the regulations of the Arbeitsagentur and obeyed several constraints respective to
each warehouse.

Tools: Or-tools, Python, Postgresql, Osrm

Generate 3D stl shapes from 2D images.
Generate and pack different shapes of cabinets, fridges and kitchen equipments based on
measurements.
Generate CAD layouts and render the result in 3D for humans to view.

Read stl files of foot 3D laser scan, fix and make mesh watertight.
Detect the metatarsal bones and the heel, then generate insole such that it obeys the doctor’s
prescription as well.

Tools: Pyvista, vtk, pymeshlab, scipy.

Use multiple reinforcement learning techniques to make a network handover protocol
“smarter”

Tools: Gymnasium, stable_baselines3, PPO, seaborn.

Transcribe audio into text.
Use CoT to do mental health analysis alongisde TF-IDF + GloVe and
BERT then generate a controlled response.

Tools: Whisper, COT, BERT, Qwen2.5, GPT-4, LLM.

Develop multiple algorithms for semantic analysis of investors’ proxy voting data and
drawing multiple conclusions over them.

Develop a semantic segmentation algorithm for voting proposals to infer a company’s
policies and insure they are inline with their public propaganda.

Tools: Python, Pandas, Spacy, NLP, sentence_transformers, BERT, LLM