Horst User

Project: Fraud Prevention using Voice, Image and Text
Since 2023
- Defined and implemented AI solutions to detect fraud in theoretical and practical driver's license tests
- Voice, image, and text data handling and analysis
- Applied object detection and Image Processing pipelines for anti-cheating solution
- Applied LLM-based prompt engineering to automate document validation at scale

Technologi

Project: Flexi Tutor AI Backbone
2022 - 2023 (1 year)
- Built an early GPT-4-powered production solution using pre-release access to the model
- Built the AI backbone for a student tutoring platform
- Use of RAG architecture and LLM prompt engineering
- Questions recognition and classification
- Implemented caching and semantic search to reduce LLM latency

Technologies: Python · TorchServe · Py

Activities:
- Use of AI techniques applied to solve image recognition problems
- Reduce of global costs with 3rd parties vendors with ML models
- Improvement of the company's app user experience through ML models
- Creation of Model Monitoring algorithms in the company's data platform
- Creation of Data Visualization dashboards and alerts to identify the need for retraining credit models
-

Clients: Vivo, Latam, SalesForce, Ambev, Unilever, Sulamérica

Activities:
- Lead teams in research and development of AI techniques
- Use of AI to solve problems like image recognition, OCR and NLP
- Creation of hybrid Image/Text Classifiers
- Creation of Named Entity Recognition (NER) models with BERT
- Use of Deep Learning and Transfer Learning for medical images processing
- Use Data S

Activities:
- Member of Aura Project - The first chatbot in Brazil powered with BigData insights
- Created anomaly detection models to identify unwanted data usage for customers
- Created geo-localization models to identify and find 3G antennas without ID
- Creation of massive failure detection model to identify failures in the wired network
- Family clustering using different data sources of unst

Activities:
- Artificial Intelligence and Machine Learning in Python
- One of the first Brazilian companies to use DeepLearning.
- Use of Caffe framework to detect home appliances images from e-commerce pages
- C++ for Image recognition
- Extract information from unstructured data like images and texts from e-commerce
- Low-Vision image processing for matching images regardless of rotatio

Automating SAP Processes with Robust, High-Performance, Zero-Error Scripts. Java Development.

MBA in Engineering and Product Management from the Polytechnic School of the University of São Paulo.