/@scottaskinosie
Loading your experience...
Preparing something amazing for you
7.6
@scottaskinosie
AI Engineer & Open Source Contributor
AI Fluency Score
7.6/10
Assessed 12/5/2025
Velocity
Generated 12/5/2025
Scott Askinosie is a data scientist and ML practitioner with two decades of mathematical modeling experience who builds production-grade AI systems with unusual rigor. He architects RAG evaluation frameworks with recall metrics, creates adversarial test sets with domain experts, and has developed his own benchmarks for catching subtle model failures—like discovering training biases others miss entirely.
These capabilities position him to lead AI adoption where reliability actually matters: building systems teams can trust, training others to work effectively with AI, and bridging the gap between prototype and production. His combination of transformer-level technical depth, professional teaching experience, and hands-on implementation is genuinely rare—and his learning velocity suggests his most impactful work is still ahead.
-Designed and evaluated multimodal content understanding systems leveraging GPT-4 Vision and vector embeddings to automatically extract and interpret technical metadata from text, diagrams, charts, and maps within engineering documentation, enabling AI-assisted content discovery workflows -Developed quality evaluation frameworks for RAG systems using LLM-as-a-judge methodologies to assess retrieval accuracy, content relevance, and metadata extraction quality across 500+ technical documents -Built and deployed production-ready agentic frameworks integrating LlamaIndex hierarchical parsing, vector search, and Model Context Protocol patterns, demonstrating scalable architecture patterns adopted by ecosystem partners -Collaborated with cross-functional product and engineering teams to design learning paths and documentation that reduced developer time-to-first-value by 40%, directly supporting feature launches and developer adoption
-Developed and deployed computer vision models for real-time food identification and volume estimation, transitioning from third-party to in-house ML solutions and delivering $2M+ in cost savings -Architected multimodal AI pipeline integrating OpenAI Vision API with production mobile application, processing 10K+ daily image classifications with 94% accuracy -Built semantic search system using Azure Prompt Flow by vectorizing Cosmos DB data, enabling context-aware query responses and improving user satisfaction scores by 28%
-Developed transformer-based NLP classification systems to categorize student feedback and dropout reasons from unstructured text, identifying 15+ distinct categories that informed targeted retention interventions and reduced student drops by 4x -Built production semantic classification models using topic modeling and BERT-based approaches to analyze Net Promoter Score responses, achieving 89% classification accuracy across 5 critical dissatisfaction areas -Conducted large-scale quantitative analysis on 487,826 student-term records demonstrating communication strategy impacts, finding call-heavy approaches increased promoter rates by 7.2 percentage points and directly informing institutional policy changes -Designed A/B tests and statistical analyses to evaluate intervention effectiveness, collaborating with non-technical stakeholders to translate findings into actionable strategies