Samuel Davis

Neil

Architected configurable LLM-based extraction agents with dynamic pipelines and prompt/object generation for scalable POC-to-prod deployment. Built a multi-agent LLM routing system for chat. Implemented MCP across Django APIs and dashboards. Fine-tuned SmolDocling for structured invoice extraction (bboxes). Added async logging and evaluation infra for GenAI agents.

Developed an invoice extraction agent using VLMs with dynamic prompts and built multi-agent intelligent chat routing and accuracy APIs, enabling configurable pipelines and production evaluation. Integrated Django APIs/dashboard, async logging, and data-prep/training pipelines to deliver production-ready ML services.

Qdox (Product under AWS BU)

Mentored MLEs across model development and post-processing. Built production ML systems on AWS (SageMaker, Lambda, SQS, DynamoDB, RDS). Integrated OCR (Textract, Tesseract, EasyOCR) used in ~100% of engagements. Developed GenAI features (RAG, chatbots, SQL) and led Auto-Tagging; evaluated Bedrock models for POCs.

Analytics for Coach

Accelerated the development of Action Recognition module along with latency & accuracy analysis with various backbone architectures.
Enhanced the efficiency of the Number Recognition module by approximately 2-3 times.

Accelerated development of action and number recognition modules, improving latency and accuracy and increasing number recognition efficiency by ~2-3x. Implemented models and backbones for sports analytics and tracking.

Intelligent Media Archival System

Implemented Tracking solution for players and optimized it for sports usecases. This played a crucial role in moving the solution to production.
Optimized GPU utilization and implemented FAISS to achieve a remarkable ~300% reduction in latency for Face Detection & Recognition modules.

Generated synthetic dataset utilizing Unreal Engine and UnrealCV for Object Detection, Instance Segmentation, and Object Tracking tasks.
Training and evaluated different models using diverse combinations of real and synthetic datasets.

Generated synthetic datasets with Unreal Engine and UnrealCV for detection, segmentation, and tracking tasks and evaluated models combining real and synthetic data to improve training outcomes.

Grade: 8.97