Architecture
- Security requirements
- Threat considerations
- Trust boundaries
- Information flows
- Permission model
Security & privacy
Where practical, Kepler26 builds AI workflows around the enterprise systems, permissions, and governance your organization already trusts.
Secure technical delivery backed by an ISO 27001-certified engineering organization.Kepler26's technical delivery is supported by our technical partner, a secure software engineering and cybersecurity organization.
Technical delivery partner
Kepler26's technical delivery is supported by our technical partner, an ISO 27001-certified software engineering and cybersecurity organization specializing in secure enterprise solutions for regulated industries.
The partnership adds software engineering, solution architecture, secure AI development, cloud, application security, DevSecOps, cybersecurity and identity and access-management depth. Kepler26 remains the client's primary consulting and delivery relationship.
Discuss technical deliveryPreferred architecture
Our default question isn't how to move your data into another platform. It's how much we can accomplish inside the environment you've already approved.
Client controls
Kepler26 provides
Feasibility depends on the client's licensed capabilities, permissions, approved providers, integration options, and governance requirements.
Implementation models
We start with the lowest-complexity architecture that can support the workflow, information boundary, and level of control required.
01 / Client-native
Kepler26 builds the workflow primarily inside technology already owned and governed by the client.
Why start here
Secure software development
For custom applications and integrations, security is considered from architecture through deployment and operation. The controls included in an engagement depend on the system, risk and agreed scope.
Specialist capability
Not every engagement needs a separate cybersecurity workstream. Kepler26 can bring the appropriate specialists into architecture, engineering and review when the information, integration or risk profile requires it.
Information boundaries
Connecting a model to an entire repository because it is technically possible usually creates more exposure, more noise, and weaker retrieval.
Better context boundaries can improve both security and AI performance.
Least privilege
Project-specific, folder-specific, read-only, time-limited, and role-based access can reduce the consequences of a mistake while making the workflow easier to understand and review.
Entire company repository
Relevant brand and project information
Agent autonomy
Autonomy should be assigned task by task. The more consequential the action, the clearer the human approval point should become.
Bounded automation lets AI perform a clearly defined, inspectable task with approved inputs and rules while people review exceptions and outcomes.
A person should normally review work before it changes an external, important, or regulated state.
AI can prepare evidence and checks. It should not quietly become the accountable decision-maker.
AI can accelerate review. It should not quietly become the accountable decision-maker.
Data classification
This example is a conversation tool, not a universal policy. Kepler26 adapts workflow boundaries to each client's classifications and governance.
Examples
Example handling
Broadest approved AI usage, subject to source quality and intellectual-property requirements.PHI and PII
Literature surveillance, congress monitoring, competitive intelligence, claims support, annotation, MLR preparation, brand planning, knowledge management, and content development often do not inherently require identifiable patient information.
Client isolation
Custom systems should prevent one client's documents, embeddings, prompts, outputs, logs, databases, or agent memory from becoming available to another client.
Safer prototyping
Where practical, early development can use synthetic documents, anonymized examples, public prescribing information, published research, fictional brands, or fabricated test data. Production can then move into the client's approved environment.
Clients do not necessarily need to provide large confidential datasets just to begin.Traceability
Appropriate workflows should make it possible to understand who initiated an action, what it accessed, when it ran, what it generated, which system it used, whether review occurred, and what final action was taken.
We design around the observability available in the chosen client technology stack.Provider governance
An approved-provider framework turns abstract policy into practical architecture decisions before a workflow reaches production.
Kepler26 can help translate these requirements into workflow, provider, integration, permission, and review decisions.
Permission readiness
Connecting AI to SharePoint, OneDrive, Teams, or another enterprise repository can make existing information dramatically easier to discover. That makes access boundaries part of AI readiness.
Identify high-value workflows, determine the information each one needs, assess available systems, and recommend a practical implementation architecture.
Before development
Give business, technology, privacy, security, legal, and procurement stakeholders a concrete architecture to evaluate before implementation.
How we approach our own operations
For this public website, analytics stays off until a visitor opts in, gated resources use passwordless access, application secrets are supplied through deployment environment variables, and the contact form asks visitors not to send confidential information.
Those website practices are not certifications and do not define a client implementation. Engagement-specific providers, access, retention, review, and documentation requirements are agreed for the work being designed.
Questions stakeholders ask
Our technical partner is ISO 27001 certified. The certification applies to our technical partner and the scope of its certified information-security management system. Kepler26 does not represent the Mody & Associates or Kepler26 legal entity as independently ISO 27001 certified.
Kepler26 leads the engagement and assembles the technical team required for the solution. Depending on the project, that can include software and AI engineers, solution architects, cloud specialists, cybersecurity specialists, DevSecOps engineers, and identity and access-management experts from our technical delivery organization.
No. Kepler26 works from workflow diagnosis and solution architecture through software development, integration, secure deployment, adoption, and ongoing optimization. The work can start small without ending at a recommendation or prototype.
Yes. The delivery model can scale from a focused workflow automation to a larger custom system or enterprise integration. Scope, architecture, staffing, and delivery stages are defined around the implementation rather than forcing every project into the same team structure.
Kepler26 remains the primary engagement relationship and coordinates the technical resources required for delivery. Specialists from our technical partner can participate directly in architecture, security, and implementation discussions when useful while Kepler26 retains solution ownership and client leadership.
Data location is determined by the approved architecture for the engagement. Where practical, information remains in client-controlled systems, regions, providers, and access boundaries. Team location and data-processing architecture are separate considerations and should not be treated as the same decision.
Not necessarily, and often no. Where practical, Kepler26 designs workflows around systems the organization already controls, such as Microsoft 365, SharePoint, approved AI platforms, cloud environments, and internal knowledge repositories.
Not always. One of our first questions is how much can be accomplished using technology the organization already owns and has approved.
When the intended workflow, available capabilities, client approval, and appropriate access support it, workflows can be designed around Microsoft 365, SharePoint, OneDrive, Teams, Copilot, Azure, and related systems. Feasibility is confirmed for each implementation.
Where the organization's plan, permissions, available integrations, and governance policies support the intended workflow, Kepler26 can design around approved enterprise AI environments.
No. We generally prefer the opposite approach: determine the minimum information required for the workflow and limit access accordingly.
AI can accelerate quality checks, evidence review, consistency checks, annotation, claims analysis, and preparation for review. Final approval decisions should remain with appropriately authorized human reviewers.
If a workflow doesn't require restricted personal information, it should not receive it. Workflows involving PHI, PII, or other restricted information require additional architecture and governance review.
Kepler26 does not design client engagements around pooling confidential client information into a shared training dataset. Data handling requirements, model-provider terms, retention, and permitted uses should be explicitly defined for each implementation.
Absolutely. Kepler26 expects security, privacy, IT, legal, or procurement stakeholders to be involved when the workflow warrants it. We can document proposed data flows, providers, permissions, integrations, and human-control points for review before production deployment.
Access should be reviewed and removed when it is no longer required. Project-specific retention or deletion requirements should be agreed as part of the implementation rather than assumed.
Bring the right people in early
Complex implementations often involve IT, information security, enterprise architecture, procurement, legal, privacy, compliance, medical, regulatory and MLR stakeholders. Kepler26 can document and walk those teams through the workflow, data sources, providers, access model, AI boundaries and deployment architecture before production.