About Immunai

Immunai is a biotechnology company founded in 2019 that applies AI and single-cell biology to map and decode the immune system. The company’s platform combines AMICA, a clinically annotated single-cell multi-omics immunology knowledgebase, with the IDE engine (Immunology Discovery Engine) to generate actionable insights that accelerate therapeutic discovery, optimize drug development, and improve patient outcomes. 

‍

Challenges

Immunai runs complex, research-driven workloads that require high-performance cloud infrastructure, strong reliability, and tight cost control. Key challenges included:

• Kubernetes and GKE performance for scientific workflows: Autoscaling and scheduling issues affected JupyterHub workloads, alongside slow image spawn times that impacted researcher productivity. 
• High-cost GPU compute and capacity planning: Managing shared A100 80GB GPU reservations across multiple projects required continuous optimization, pricing strategy, and scenario modeling. 
• FinOps visibility and billing confidence: Internal stakeholders need SKU-level cost transparency, accurate invoice validation, and dependable cost attribution across teams, namespaces, and workloads. 
• Networking and platform reliability: DNS design (public and private split-horizon), load balancer modernization, and production database incidents required fast, specialized support. 

‍

Solutions:
Commit operated as Immunai’s Technical Account Management partner in an ongoing engagement model, combining hands-on troubleshooting, optimization, and proactive delivery. Core actions included:

• GKE and JupyterHub optimization: Troubleshot autoscaler and scheduling behavior, improved image streaming, and delivered a caching approach that reduced JupyterHub spawn time from 15–30 seconds to under 5 seconds. Commit also supported load balancer migration to a Regional Internal Application Load Balancer and enabled private environment ingress configurations. 
• GPU cost and reservation strategy: Managed and optimized A100 GPU reservations and led discount negotiations, including 1-year CUD options with materially improved rates. Commit provided detailed cost simulations to support decision-making across pricing scenarios. 
• FinOps foundation in BigQuery: Established BigQuery billing export as the source of truth, developed a custom Reservation Dashboard to attribute costs by user, namespace, and workload, and addressed label propagation gaps that impacted attribution (PVCs and boot disks). 
• Rapid incident response and specialist access: During critical database incidents, Commit coordinated specialized DBA support in under an hour, including outside standard business hours. 
• Added-value engineering: Delivered production-ready enablement such as a BigQuery-to-Slack agent integration, including code, documentation, Terraform modules, and Kubernetes manifests. Commit also coordinated joint roadmap sessions with Google to evaluate emerging offerings such as BigQuery Agents, Gemini Enterprise, and next-generation GPU options. 

‍

Results

• Material GPU cost reduction: In a documented simulation, the monthly cost of an a2-ultragpu-8g instance decreased from approximately $29,462 (list price) to approximately $15,795 after negotiated discounts, representing roughly 46% savings. In addition, 1-year CUDs with up to 52% discount on GPU SKUs generated savings of over $10,000 per month per instance. 
• Improved researcher experience and platform efficiency: JupyterHub image spawn time improved from 15–30 seconds to under 5 seconds, reducing friction for day-to-day scientific workflows. 
• Higher confidence in cloud spend and allocation: BigQuery-based cost visibility and a custom Reservation Dashboard enabled clearer attribution and faster resolution of billing questions, supporting finance stakeholders and operational teams with consistent data. 
• Faster resolution for business-critical issues: A dedicated operating cadence and a responsive Slack-based collaboration model improved time-to-troubleshoot across infrastructure, networking, and database incidents. 
• Stronger partnership model with Google Cloud: Commit acted as the technical arm in a three-way collaboration with Immunai and Google, helping convert roadmap opportunities, discount programs, and product access into implemented outcomes.

‍

Customer Quote: 

“Commit was there as our business grew and our needs evolved. They helped us improve cloud cost visibility, optimize Kubernetes, scale major workloads, and respond quickly to incidents. Their technical expertise and personal commitment made a real difference.”

‍Guy Yachdav, Sr. Director of Software Engineering, Immunai

‍