DIRECTV for Business Product Development Intern

What you'd do

• Develop artificial intelligence and machine learning tools
• Contribute to the development of an AI-powered media intelligence platform that analyzes live and recorded video content.
• Work with video metadata, closed captions, speech-to-text transcripts, and AI/ML models to identify topics, summarize content, detect relevant events, and enable natural-language search across video streams.
• Evaluating AI models, developing data-processing pipelines, experimenting with embeddings and semantic search, and building prototype tools or dashboards to demonstrate new media intelligence use cases.
• Duration: 10-week internship (3-5 days per week in the Clarksburg, MD office)
• Program Dates: 6/7/27 – 8/13/27 or 6/14/27 – 8/20/27
• Eligibility: Must have full U.S. work authorization now and in the future
• This position can earn $25 - $34 per hour, not to mention all the other amazing rewards that working at DIRECTV offers. (Hourly rate based on academic status)

What they want

• Students pursuing a four-year undergraduate or advanced degree in Artificial Intelligence, Computer Science, Software Engineering, Information Technology, or Data Science.
• Required: Proficiency in Python and its ML ecosystem (NumPy and pandas), with the ability to write reproducible training and evaluation scripts.
• Assist in selecting the appropriate machine learning approach (supervised, unsupervised, deep learning, or reinforcement learning) based on problem type and data availability
• Support the end-to-end model development lifecycle, including data collection, preprocessing, feature extraction and selection, training, and validation
• Build and benchmark baseline models in Python, including linear regression, decision trees, and support vector machines, for classification and regression tasks
• Operating Systems and tools: Linux and/or Android, Claude Code and OpenAI API
• Apply unsupervised learning techniques such as clustering, principal component analysis, and association rule mining to explore unlabeled datasets
• Identify and address overfitting and underfitting, applying bias–variance trade-off analysis during model tuning and selection
• Evaluate model performance using appropriate metrics (accuracy, precision, recall, mean squared error) and recommend evaluation criteria suited to the problem domain
• Experiment with ensemble methods (bagging, boosting, random forests) to improve prediction accuracy and generalization over single-model baselines