Bryan Dickson

In my most recent role as a Senior Machine Learning Engineer at TFORM in Irving, TX, where I worked remotely on a contract basis from June 2022 to January 2024, I led the development of computer vision applications. My responsibilities included implementing object detection and segmentation algorithms using CNNs and VIT models. I collaborated with cross-functional teams to integrate computer vision solutions into client projects, enhancing product features and user experiences. Additionally, I played a pivotal role in providing strategic and technical direction, mentoring team members, and ensuring the delivery of actionable insights to customers.

Led the development of computer vision applications using state-of-the-art techniques, including Convolution Neural Networks (CNNs) such as Mobile-net and Vision Transformer (VIT) models.
Implemented object detection and segmentation algorithms, enhancing accuracy and efficiency in real-time object recognition tasks.
Collaborated with cross-functional teams to integrate computer vision solutions into client projects, resulting in improved product features and user experiences.
Provided strategic and technical direction, offer mentoring and career growth at all levels, and ensure that we are rapidly delivering new actionable insights and data to our customers.
Implemented deep learning models and numerical Computation with the help of data flow graphs using Tensor Flow.
Created a new way to identify automatically samples which are not helping the model to learn (data-centric approach).
Created a new format to be able to access the dataset in a random way as fast as possible.
Balanced dataset by smartly weighting the samples.
Recorded multiple metrics on dataset per samples and displaying them in 2D/3D (+ colors).
Recorded embeddings as a metric to identify a data shift.
Made the trained model for serving by quantization, pruning and distillation
Worked and extracted data from various database sources like Oracle, SQL Server, DB2, and MongoDB.
Engineered NLP models for various applications, leveraging libraries such as NLTK, spaCy, and Huggingface Transformers.
Implemented advanced NLP techniques, including Named Entity Recognition (NER), sentiment analysis, and text summarization, providing valuable insights from unstructured data.
Worked extensively with Large Language Models (LLMs) such as OpenAI's ChatGPT and Huggingface's models, utilizing them for diverse tasks, from conversational agents to content generation.
Developed and deployed machine learning models for diverse applications, leveraging algorithms such as regression, classification, and clustering to extract valuable insights and predictions.
Extensive experience in integrating machine learning models into RESTful APIs, ensuring seamless communication between applications and enabling real-time predictions and data-driven decision-making.
Proficient in leveraging Azure's cloud services for machine learning, including Azure Machine Learning Studio, Azure Databricks, and Azure ML Services, to build scalable and efficient end-to-end ML workflows.
Implemented robust data processing pipelines and feature engineering techniques to prepare and preprocess datasets, enhancing the performance and accuracy of machine learning models.
Successfully deployed machine learning models on Azure, establishing monitoring mechanisms to track model performance, ensure reliability, and facilitate continuous improvement through iterative updates.

Led Deep Learning research and integrating an end-to-end data pipeline from inception into our main product, Interactive AI/Pickasso (papers published on optimizing models with padding valid + random optimized accessed, presented at SEG 2019).
Led research for Data Centric new methods to improve current results (smart data sampling, cleaning, embeddings on the fly) and integrated all of these results into backend directly (patents approvals in progress).
Developed innovative solutions that combined computer vision and NLP capabilities. For instance, implemented Visual Question Answering (VQA) systems where images were analyzed using computer vision, and questions about the images were answered using NLP models.
Supervised a 10-man team that built a machine learning system, which makes weekly interest rate estimates for a $1B tech start-up; increased consumer spending by 94% in the first year of the project.
Spearheaded research and development projects to explore synergies between image analysis and textual data, enabling cross-modal analysis for complex tasks like image captioning and sentiment-aware image classification.
Built and deployed an end-to-end Fraud detection solution, including NLP + anomaly detection models and an automated pipeline, which saves several $M annually by flagging large-scale fraud events weeks earlier than the previous process.
Built NLP-based solutions for risk mitigation and fraud detection (topic modeling (LDA), text summarization, named entity recognition (NER) - tech stack: BERT, gensim, nltk, spacy).
Built time series forecasting models (time series as supervised learning models (with random forest and xgboost).
Developed containerized (Docker) deployments of ML frameworks (e.g. Spark, H2O, distributed TensorFlow, Airflow) to a Kubernetes cluster.
Developed machine learning models to improve venues’ quality (Lat/Lng accuracy, Venue Category etc.) using supervised / unsupervised learning algorithms.
Managed venue data pipeline. Create, manage, and optimize various data pipelines in Scala and Python, using Luigi and Spark.
Researched and implemented several computer vision applications for Hearst Magazines: image labelling, find-images-by-image, magazine face recognition.
Developed Smart News application to help editors to decide what to write about using convolutional Neural Network based on historical data.
Managed and develop Hearst Digital Management Platform to perform audience analysis, segment analysis.
Created a framework to auto-tune our parameter space to avoid our users to tweak IAI parameters manually.
Planned following business guidelines, reporting to CTO and board of investors/partners.

Collaborated with Business Analysts, SMEs across departments to gather business requirements, and identify workable items for further development.
Partnered with ETL developers to ensure that data is well cleaned, and the data warehouse is up to date for reporting purposes by Pig.
Implemented and deployed NLP algorithms such as question answering, information retrieval and text classification based on transformers.
Utilized visualization tools like Apache Zeppelin, Matplotlib, and Tableau to create interactive and informative visualizations from both computer vision and NLP analysis results.
Translated complex technical findings into comprehensive reports and dashboards, enabling stakeholders to understand the impact and potential applications of the developed models and algorithms.
Implemented Text Retrieval using BI-Encoder and Cross-Encoder.
Implemented Data augmentation for question-answering for finetuning tasks using language models, contextual word embeddings, back translation and abstractive summarization.
Deployed models to production using Docker containers and Kubernetes.
Built and deployed API’s and microservices.
Implemented of unit tests, integration tests, regression tests and smoking tests.
Researched and propose improvements to core system using recent state-of-the-art works.
Implemented and deployed algorithms to solve complex NLP problems such as question answering, text classification and text summarization.
Worked with algorithms like OCR, Sentence Bert, Text Retrieval using BI-Encoder and Cross-Encoder, Semantic Search.
Developed a chatbot which designs a Resume of the user by asking necessary questions. All the user responses were stored in the MongoDB Database.
Researched and proposed improvements to core system using recent state-of-the-art works.
Reviewed and fixed advanced bugs in core system.