Daniel Miller Email & Phone Number

● Team Lead. Developed and deployed end-to-end pipelines for large scale inference utilizing ECS, EC2 or Sagemaker.● Curated custom datasets through data collection, annotation, and advanced pre-processing techniques.● Created custom training and fine-tuning models through code reverse engineering using single GPU or multiple GPUs for diffusion, transformers and deep learning based models.● Prepared AI pipelines for digital human generation, cloned voice, text-to-text, text-to-image, image(s)-to-3d model, audio-to-text, text translation, texture mapping and virtual clothing try on.● Created a custom background removal model for various e-commerce platforms, using U-Net architecture and TensorFlow for optimized image segmentation.● Collaborated on developing a tax advisory chatbot utilizing transformers and large language models (GPT-3) for interactive, intelligent responses.● Implemented customer and product analysis pipelines in retail and industrial environments, utilizing YOLO for object detection, Deep SORT for object tracking, and custom OCR models.● Deployed applications via AWS App Runner and RDS, integrating OpenAI API and sentence embeddings for efficient resume and CV analysis.● Built an end-to-end pipeline for course content generation and multiple-choice question creation, using LangChain and OpenAI models, tailored to various kid profiles and subjects.

● Developed and led end-to-end pipelines, specializing in Machine Learning and Deep Learning models using TensorFlow, Keras, and PyTorch for Edge Device deployment (Android, iOS, Micro-controllers).● Created tailored datasets through data collection, annotation, and pre-processing techniques.● Achieved high model accuracy and efficiency through Fine Tuning, Transfer Learning, Ensembling and model optimizations.● Enhanced real-time Face Recognition accuracy to approximately 98% and achieved around 98% accuracy for Face Anti-Spoofing, using ambient light and liveliness detection.● Attained a 91% mAP for car detection from a Drone Perspective View with custom datasets and improved object localization on maps, reducing error to less than 5 meters by employing perspective transformation, photogrammetry, and feature-matching techniques.● Reduced manufactured part rejection rates by approximately 95% using a Flask-API system and integrated face recognition in noisy industrial environments with a matching accuracy of around 98%.

● Developed predictive models for customer matching and pet-sitting recommendations, enhancing matching accuracy.● Built machine learning algorithms for pet behavior analysis, identifying risk behaviors from image and video data.● Created scalable data pipelines for processing high-volume customer data with continuous feedback loops.● Deployed pet profile analysis features leveraging AWS Lambda and S3, improving processing efficiency.● Conducted A/B testing for model-driven features, enhancing user engagement metrics.