Adam Wawrzynski

Developed a backend for an NLP microservices pipeline that generates meeting notes via summarization and multiple auxiliary language tasks. Integrated pre-trained neural models for summarization, punctuation restoration, intent classification, sentiment analysis, and keyword generation across languages. Deployed transformer-based components as services and supported scalable inference.

Built a retrieval-augmented generation system to answer questions using open-source library documentation. Implemented a document retrieval pipeline and integrated it with an LLM-based assistant workflow. Focused on efficient retrieval using vector search and the orchestration layer for RAG.

Designed and trained multimodal models for ad creative evaluation and explanation generation. Prepared training data and addressed the challenge of making model explanations actionable for end users. Supported productionization and ran performance/load testing for model-backed services.

Iterated on an object detection model deployed on a robotic station with data distributions that changed over time. Performed daily error analysis, selected samples for fine-tuning, and coordinated the annotation workflow. Implemented fine-tuning and evaluation loops to improve detection quality under real operational conditions.

Worked with a client to translate business ideas into ML tasks for technical documentation processing. Developed a research-driven PoC focused on extracting and structuring information from documents using OCR and detection-based approaches. Assessed feasibility, limitations, and efficiency of candidate ML solutions with an emphasis on practical outcomes.