Jen Gillaspie

• Managed data labeling and generation projects from initiation to completion, ensuring the highest quality.

• Assigned responsibilities to team members to ensure the best resource use and tracked project milestones and deliverables.

• Directly managed and led the large community of contractors, acting as an advocate for both the company and the independent contractor.

• Delivered the highest quality standard in the industry and provided training to ensure accuracy across the board.

• Facilitated cross-functional collaboration and communicated wins and challenges with suggestions for improvement.

• Led LLM projects requiring human feedback, including red teaming, preference ranking, prompt evaluation, and multimodal testing.

• Annotated and prepared datasets, fine-tuned over 15,000 units for large language model training while adhering to safety, ethics, and instruction alignment.

• Crafted comprehensive project guidelines and created golden annotated datasets that elevated labeling and model evaluation accuracy across teams.

• Facilitated onboarding and training for over 105 new LLM specialists over 3 months, authoring training materials and best practice documentation.

• Conducted direct QA and feedback cycles to improve model performance and annotator alignment, receiving an award for "human-centered leadership and high performance."

• Executed LLM fine-tuning projects in supervised and unsupervised training pipelines for prompt-response datasets.

• Consistently selected to work on urgent Generative AI pilots, provide detailed feedback documentation, and maintain flexibility with changing client guidelines.

• Evaluated AI-generated content for safety, accuracy, toxicity, helpfulness, appropriateness, tone, and style.

• Performed quality assurance reviews across crowd annotation teams to ensure 100% data integrity.

• Supported diverse AI tasks, including image labeling, transcription, speech/video-to-text, prompt creation, and social media platform testing.

• Created cleverly worded prompts to assess model safety via jailbreaking techniques.