Compute
Sized per workload- GPU inferenceNVIDIA GPU instances in your cloud or on-premises
- ServicesContainerized and autoscaled
- OrchestrationKubernetes (AKS, EKS, GKE) or VMs
- ScalingBy queue depth and latency
ATG TechPlatform
The production foundation for enterprise AI.
A platform we deploy inside your cloud or data center: connectors, retrieval, models, agents, guardrails, and observability, assembled and operated as one governed system.
Select a layer to see what it does
Runs inside your boundaryYour cloud tenant, a private VPC, or your own data center.
Model-agnosticHosted APIs or open-weight models, chosen per task.
Traceable answersEvery response links back to the sources it used.
Built, co-run, or fully runPick the operating model that fits your team.
Deployment
Deployed where your data already lives.
We stand the platform up inside the environment you control: your Azure, AWS, or Google Cloud tenant, a private VPC, or your own data center. Your data stays inside the boundary your security team already governs.
- Infrastructure as code, reviewed with your cloud team
- Separate development, staging, and production environments
- GPU capacity sized to your workloads and scaled with demand
Your environment · Azure, AWS, Google Cloud, or your data center
Private network
ATG TechPlatform
- L1Integration
- L2Knowledge
- L3Model gateway
- L4Agent orchestration
- L5Guardrails
- L6Observability
SharePoint
Databases
CRM · EHR
Your people sign in with SSOPublic internet egress off
Security
Private by default.
Model calls, retrieval, and agent actions travel over private networking, authenticate through your identity provider, and are logged for audit.
EmployeeSigns in once
Identity providerEntra ID · Okta · SAML
Model gatewayPolicy and budget check
Permission filterOnly documents this user may see
AnswerCited, logged, auditable
- Private endpoints, with public egress restricted or disabled
- Encryption with customer-managed keys
- Least-privilege credentials for every agent tool
Operations
Operated with full visibility.
See what every agent did, which sources it used, what it cost, and how accurate it was, so quality improves on evidence rather than guesswork.
- Traces for every step, searchable by user and application
- Scheduled evaluation runs against approved test sets
- Alerts for quality drift, stale content, latency, and spend
Example traceclaims assistant · one request
receivedanswered
Receive questionclaims analyst · via Teams
Retrieve & rerank4 sources, permission-filtered
Guardrails2 fields masked before the model call
Model callanswer drafted from sources only
Citation check3 of 3 claims verified
Log & returnaudit entry written
Faithfulness passedCost recordedNo approval needed
Technical specifications
What's under the hood.
Every component is chosen per engagement and documented in your repositories. These are the building blocks we deploy most often.
Data & storage
Stays in your account- Vector indexpgvector, Azure AI Search, OpenSearch, or Pinecone
- DocumentsObject storage in your account
- MetadataPostgreSQL
- BackupsScheduled, encrypted, restore-tested
Models
Chosen per task- HostedAzure OpenAI, AWS Bedrock, Vertex AI
- Open-weightLlama, Mistral, and others on your GPUs
- EmbeddingsHosted or self-hosted
- RoutingPer task, with automatic fallback
Network
Private by default- ConnectivityPrivate endpoints, VNet or VPC peering
- EgressRestricted or fully disabled
- EdgeWAF and rate limiting
- RegionPinned to the region you choose
Security
Identity-aware- IdentitySSO via Entra ID, Okta, or SAML
- AccessRBAC and ABAC
- EncryptionTLS 1.2+, AES-256, customer-managed keys
- AuditImmutable request and action logs
Observability
Measured continuously- TracingOpenTelemetry-compatible
- EvaluationFaithfulness, relevance, citation accuracy
- DashboardsQuality, usage, latency, and cost
- AlertingEmail, Teams, Slack, or PagerDuty
Deployment options
Four ways to run it.
Pick a model to see which components run inside your environment and which, if any, sit outside it.
Built for the agents your teams actually need.
Five kinds of work we deploy most often, and the kind of request each one handles.
Knowledge assistants
Answers from policies, manuals, and records, with citations people can check.
Someone asks“What changed in the 2026 travel policy?”
Document and back-office agents
Claims, invoices, applications, and forms processed with a person approving each exception.
Someone asks“Process today's claims and flag anything that's missing.”
Voice agents
Phone scheduling, intake, and status calls grounded in your systems.
Someone asks“I need to move my Thursday appointment.”
Analytics copilots
Plain-English questions answered from governed data, with the query shown.
Someone asks“Which regions missed their Q3 target, and why?”
Customer-facing chat
Support and self-service that stays on policy and hands off cleanly.
Someone asks“Where's my order, and can I change the address?”
Built, co-run, or fully run.
However it's operated, you own the platform, the code, and the data. The difference is who carries the pager.
| Responsibility | Build & hand overYour team operates it from day one | Co-managedWe operate alongside your team | Fully managedWe run it; you own it |
|---|---|---|---|
| Architecture, build, and launch | |||
| Infrastructure as code in your repositories | |||
| Runbooks and team training | |||
| Monitoring and incident response | — | ||
| Evaluation runs and quality reports | |||
| Model and connector upgrades | — | ||
| Cost optimization reviews | — |
Build & hand overYour team operates it from day one
- Architecture, build, and launch
- Infrastructure as code in your repositories
- Runbooks and team training
- —Monitoring and incident response
- Evaluation runs and quality reports
- —Model and connector upgrades
- —Cost optimization reviews
Co-managedWe operate alongside your team
- Architecture, build, and launch
- Infrastructure as code in your repositories
- Runbooks and team training
- Monitoring and incident response
- Evaluation runs and quality reports
- Model and connector upgrades
- Cost optimization reviews
Fully managedWe run it; you own it
- Architecture, build, and launch
- Infrastructure as code in your repositories
- Runbooks and team training
- Monitoring and incident response
- Evaluation runs and quality reports
- Model and connector upgrades
- Cost optimization reviews
ATGShared with your team— Your team
Compliance readiness
Designed for regulated data.
The platform is built to support the frameworks your auditors care about. Certification and compliance remain with your organization; we provide the controls, documentation, and evidence to support them.
| Framework | Controls we implement | Evidence you receive |
|---|---|---|
| HIPAAProtected health information | PHI masking, role-based access, audit logging, and deployment on HIPAA-eligible cloud services | Access and audit reports |
| SOC 2Security, availability, confidentiality | Access reviews, change control through infrastructure as code, encryption, and centralized logs | Control narratives and log exports |
| GDPR & CCPAPersonal data rights | Data inventory, retention rules, deletion workflows, and region pinning | Inputs for records of processing |
| FERPAStudent education records | Role-based, minimum-necessary retrieval and per-record audit trails | Access logs by user and record |
| GLBACustomer financial information | Encryption, access control, isolation from other workloads, and monitoring | Safeguards documentation |
Plan the platform your agents will run on.
Tell us where your data lives and what you want AI to do with it. We'll come back with an architecture and a phased plan.