SAMUEL APOYA

Research assistant with a strong foundation in machine learning and data analysis, honed at the Insite Lab. Proven ability to develop innovative solutions, such as context-aware image descriptions for blind users. Skilled in Python and collaboration, driving impactful projects that enhance accessibility and user experience.

• Assistant Program Coordinator & SAT Math Tutor, Young Achievers Foundation Ghana, 06/24, 08/24, Coordinated college preparatory programming for 40+ high school students, managing curriculum planning, assessments, and scheduling., Provided individualized and group SAT Math tutoring, contributing to average score improvements of 80-100 points., Supported students through college application planning and academic advising.
• Club Member, Colby Hackers, Waterville, ME, 09/24, present, Participate in hackathons and mentor student teams on ML project design, technical implementation, and presentation strategies, contributing to multiple award-winning projects including QCare at HackPrinceton.

• Research Assistant, Davis Institute for Artificial Intelligence, 10/24, 05/25, Investigated why Colby's campus chatbot produced confident but fabricated answers to student queries, framing the problem as a retrieval-augmented generation research question under faculty guidance., Built an end-to-end RAG pipeline by ingesting institutional documents including handbooks, catalogs, and schedules, implementing semantic chunking strategies, and indexing passages using FAISS for efficient vector search., Explored how different retrieval strategies affect answer reliability by testing BM25 lexical matching, Sentence Transformers semantic search, and hybrid combinations, tracking whether generated responses could be traced back to source documents., Developed interactive web application to teach transformer attention mechanisms, supporting broader AI literacy efforts at Colby.
• Research Assistant, The Insite Lab, Colby College Computer Science Department, Waterville, ME, 05/25, present, Collaborate with faculty to develop spatially grounded image descriptions for blind and low-vision (BLV) museum visitors using multimodal machine learning for accessibility., Fine-tune LLaMA using parameter-efficient LoRA adapters and align CLIP visual embeddings with spatial language representations via contrastive loss and bounding box supervision., Generate context-aware descriptions that encode spatial relationships rather than object labels alone, enabling BLV users to construct accurate mental maps of artwork layouts., Build a curator-facing web application with React and MongoDB for generating and versioning accessibility descriptions; conducting ongoing BLV user studies to validate system effectiveness toward publication.
• Research Assistant, Colby College & Bigelow Laboratory for Ocean Sciences, Waterville, ME, 02/23, present, Analyzed time-series environmental data for sustainable seaweed aquaculture supporting climate mitigation., Cleaned and processed noisy sensor data on temperature, nutrients, and yields using Pandas and NumPy, handling missing values, irregular sampling, and sensor drift., Trained a temporal convolutional network (TCN) in PyTorch to predict growth yields, benchmarking against LSTM and XGBoost; models informed cultivation site selection for Maine coastal expansion.
• Teaching Assistant, Colby Computer Science Department, Waterville, ME, 02/23, present, Support students in Machine Learning and Software Engineering courses through weekly TA hours and lab sessions., Assist with debugging neural network training pipelines, optimization algorithms, and models built in PyTorch and TensorFlow., Mentor students on gradient descent, backpropagation, model evaluation, and algorithmic complexity in Python and Java.