Define. Prove. Engineer. Evolve.
A repeatable process for software projects with and without AI, with an explicit decision at each transition. Each stage is only as large as the project needs.
- 1
Define.
Understand the business problem, the users, the systems involved, the requirements, the scope and how success will be measured.
Decision: Is the problem clear and the scope agreed?
- 2
Prove.
Validate the important uncertainties with a prototype or test where needed. For a straightforward software project this may simply be design validation.
Decision: Is the approach sound enough to build?
- 3
Engineer.
Build, integrate, test, deploy and prepare the agreed release.
Decision: Is the release ready for its intended use?
- 4
Evolve.
Ongoing operation, maintenance and improvements, or an agreed handoff to your team.
Decision: What should be maintained, improved or handed over?
Clear roles, a real handover, explicit acceptance.
Client collaboration
You provide an accountable sponsor, representative users, the required data access and timely decisions. Cennix owns the agreed engineering deliverables and communicates progress, open questions, dependencies and material scope changes.
Handover
Operating instructions, relevant source or access under the agreement, deployment information, known limitations, support contacts or routing, and a clear split of responsibilities. You should understand what you bought.
Acceptance
Explicit criteria tied to real tasks: answer quality or task completion, source support, authorization behavior, error handling, operating performance and the intended user experience. One universal "AI accuracy" number is not enough.
The simplest architecture that meets the requirements.
The value is the complete system and its operation. Frameworks, providers and infrastructure vendors are implementation choices, not our identity.
User experience
Web applications, internal interfaces, portals, document views, and review or approval screens.
Application logic
Business rules, APIs, task orchestration, user permissions and state management.
Intelligence
Appropriate models, prompts, retrieval, structured outputs, agent tools and evaluation.
Data and integration
Authorized data ingestion, document processing, search indexes, databases and connections to your systems.
Runtime and operation
Deployment, model access, monitoring, logs, backups where applicable, and release management.
How a system behaves when things go wrong matters as much as when they go right.
- 1.
Model selection
Models and providers are chosen on task performance, data requirements, speed, cost and deployment constraints, and compared on the project's own evaluation examples instead of assuming the largest model is the best fit.
- 2.
Evidence and failure behavior
A knowledge application distinguishes supported answers from missing information. An agent respects its action boundaries and handles failed tool calls explicitly. A decision-support application exposes its assumptions and keeps the required review step.
- 3.
Reuse with judgment
Reusable foundations, evaluation tools and deployment patterns make delivery more consistent. Reusable components stay separate from client-specific data and deliverables.
Ways in, not a fixed path.
Every proposal states the problem, users, deliverables, success criteria, assumptions, exclusions, required inputs, milestones, ownership terms and any ongoing costs.
Engagement
What it covers
Commercial structure
Project conversation
A short call about the problem, or a brief written recommendation on how to approach it. A lightweight introduction, not a full plan.
No charge
Solution Design & Consulting
Requirements, architecture, a build-versus-buy assessment, a practical roadmap and a scoped estimate. Useful on its own or before a build.
Scoped separately
Custom Business Software
CRMs, dashboards, internal tools, customer portals and integrations, built around how your teams work.
Scoped project, in milestones
Product Development
Product definition, a prototype where useful, a first usable release, the production application and later iterations.
Scoped project, in milestones
Custom AI Implementation
Knowledge systems, agents, AI features, integrations and the infrastructure to deploy and operate them.
Scoped project, with a prototype when uncertainty warrants it
Managed Hosting & Ongoing Engineering
Hosting where suitable, monitoring, maintenance, support and agreed improvements after launch. Optional: a handoff is always possible.
Optional, agreed monthly scope