Ryan Orban

Borrowed sequence alignment from bioinformatics — Needleman-Wunsch, progressive multi-sequence alignment — to align agent trajectories across 12,854 runs and locate where passing and failing runs diverge, with spectrum-based fault localization from software debugging layered on top. Alongside it, a sequential-testing harness that stops an evaluation as soon as the evidence is decisive, so you pay for the runs you need rather than a fixed number.

The same question, pointed at models: what has to be true before an LLM can touch regulated data? Built secure workspace chatbots with RBAC-aware RAG and agentic pipelines, and the verification around them — red-teaming, data governance, evals. Judged the market early rather than late, returned the remaining funds, and kept the playbooks.

A network sells work from people the client never hired, so trust is the entire product. Built the vetting and delivery standards that made 150+ independent senior ML practitioners reliable enough to sell — standardized playbooks, shorter scoping-to-delivery cycles — and served as tech lead for client work across LLM prototyping, recommender systems, and data platforms.

The same judgment, automated: ranking people against roles at scale. Built the matching stack — Elasticsearch with Learning To Rank, BERT embeddings — which improved placement conversion, established with Bayesian A/B testing rather than assumed.

Three years overland through North, Central, and South America.

Data science had no credential, so employers had no way to tell who could do the work. Built one and proved it out on the only metric that settles the question — where graduates landed. Bootstrapped the first immersive data science program in the US with a 10-person team and no outside capital, grew revenue past $1M in year one, placed 91% of graduates at top tech firms including Tesla, Facebook, and Google, and created a curriculum format that many later programs echoed.

Galvanize acquired it, and I led the post-acquisition integration: stood up enterprise training and assessments, scaled the data science curriculum nationwide, hired and managed instructor teams, and aligned pedagogy with industry needs.

Started in distributed systems engineering, then moved to the customer side across Army, Air Force, and intelligence — buyers who grant no trust by default and have to be shown. The proof was the job: large-scale POCs demonstrating a system would hold before anyone would deploy it. Designed and deployed $100M+ clusters, translating hard infrastructure constraints into resilient systems.