Case studiesWhat engagements look like.
Representative engagements showing scope, approach and the targets we set. Detailed client references are available under NDA.
case-lender1200 × 800 px
Finance · Private LLM DeploymentRegional lender moves credit memo drafting on-prem
A regional lender wanted AI help drafting credit memos but could not send borrower data to an external API.
- Scope: Llama-family model on four L40S GPUs in the bank's data center, vLLM serving, SSO via Entra ID
- Integration: Loan origination system and document store, with row-level permissions
- Controls: Prompt and response logging to Splunk, model change approvals
0 external API callsDraft time: hours → minutes (target)Model risk documentation included
case-saas1200 × 800 px
AI Startup · GPU OptimizationSeries B SaaS cuts cost per request
An AI-native SaaS company saw its inference bill grow faster than revenue as usage scaled.
- Diagnosis: Utilization audit showed low batching and a single large model serving every request
- Changes: Model routing, AWQ int4 quantization, continuous batching, autoscaling
- Governance: Per-feature cost dashboards and budget alerts
Target −50% $/1M tokensp95 latency held under SLOMonthly FinOps review
case-health1200 × 800 px
Healthcare · AI Security AuditHealth system audits AI before rollout
A health system needed to approve several AI tools and had no process for assessing them.
- Inventory: Every AI tool and vendor touching PHI, including unsanctioned ones
- Testing: Red-team exercises on a clinical documentation pilot
- Output: HIPAA-mapped risk register and AI acceptable-use policy
HIPAA-mapped risk registerPolicy pack adoptedApproval process for new AI tools
Private✦Secure✦Efficient✦Yours✦On-prem✦Air-gapped✦Private✦Secure✦Efficient✦Yours✦On-prem✦Air-gapped✦
vLLM●H100●LoRA●RAG●MLOps●FinOps●Zero egress●vLLM●H100●LoRA●RAG●MLOps●FinOps●Zero egress●
Next stepWant a reference call?
We'll connect you with a client in a similar situation, under NDA.
