Independent AI Consultant: Agents and LLM Apps That Reach Production
Most New Zealand organisations have tried AI. Far fewer have it running in real operations. I design and build AI agents and LLM applications on your own data and systems, and take them from proof of concept to something your team relies on every day.
Why AI pilots stall
- The demo impressed, but it cannot answer questions about your own documents or data.
- The AI is not connected to the systems where the work actually happens.
- Staff do not trust the answers because they cannot see where they came from.
- Privacy and security concerns stop the rollout at the last step.
- Copilot licences were bought, but nobody can point to measurable value.
- Nobody owns or maintains the solution once the pilot team moves on.
What I build
Model choice is the easy part. The value comes from connecting AI to your data, your systems and your processes, and from making it reliable enough that people use it.
Custom AI agents
Agents that complete multi-step work across your systems: triaging requests, drafting documents and updating records, with human approval where it matters.
RAG on your own knowledge
AI that answers from your policies, procedures and documents, citing the source so staff can check it.
LLM applications
Purpose-built applications on Claude or OpenAI models, chosen for accuracy, cost and data sensitivity.
AI connected to your systems
Secure integration with Microsoft 365, SharePoint, databases and operational systems through APIs and connectors.
Microsoft Copilot and Copilot Studio
Getting real value from Copilot licences, and Copilot Studio agents where Microsoft is the right fit.
Evaluation and guardrails
Testing AI against real cases before launch, monitoring accuracy and cost, and keeping people in the loop on decisions that matter.
My End-to-End Approach
One senior person accountable from the first workshop to measured results. No handoffs between analyst, architect and developer, so nothing gets lost in translation. Engage me for the whole journey, or for any single stage.
-
01
Understand
Business analysis first: stakeholder workshops, current-state process mapping and the real problem behind the request.
You getProblem statement and current-state map -
02
Quantify
Size the prize: what the current way of working costs, options compared, and ROI in terms a board can approve.
You getBusiness case and prioritised options -
03
Design
Requirements, data and solution design, choosing the right answer independently: AI, automation, reporting or simply a better process.
You getRequirements and solution design -
04
Build
Hands-on delivery of AI agents, automation, Power BI and integrations, using AI-assisted engineering to show working results early.
You getWorking solution, tested on your data -
05
Embed
Change support, training, documentation and handover, so your people actually use it and can run it.
You getA solution your team owns -
06
Measure
Track the benefits against the business case and improve whatever is not delivering.
You getMeasured results
AI I have built, not just advised on
I co-founded eimprovers and built Semantic Robot, a semantic monitoring engine licensed by parental control apps including FaceAlert. I founded AIGuardU, which analyses children's AI conversations in real time for sentiment and concern patterns, with a privacy-first design that discards conversation content immediately and never stores it. I build production software with modern AI-assisted engineering, which shortens the path from idea to working prototype dramatically, and I bring 20 years inside NZ operations to decide what is worth building.
Every engagement is delivered by Luis Silva personally. No juniors, no subcontractors, and no vendor commissions.
Ways to work together
Questions
Which AI models do you use?
I am model-agnostic. I work with Claude, OpenAI models and Microsoft Copilot, and choose based on accuracy, cost and how sensitive your data is. I earn no commissions from any AI vendor.
How do you keep our data safe?
Solutions are designed so your data stays under your control: enterprise model agreements, the minimum data sent to a model, access that mirrors your existing permissions, and human review of important decisions.
We already have Microsoft 365 Copilot. Do we need custom AI?
Sometimes not. Copilot covers general productivity well. Custom agents make sense when AI must work with specialised data, operational systems or a defined business process. I will tell you honestly which you need.
How do we start?
With a 30-minute process review. If there is a strong use case, the next step is a fixed-fee proof of concept on your real data, so you see results before committing to a full build.
Can AI agents connect to our existing systems?
Yes. Agents can work with your existing systems through APIs, standard connectors and, increasingly, the Model Context Protocol (MCP), the emerging standard for connecting AI to business tools.
Tell me your biggest bottleneck
Book a 30-minute process review. No commitment and no sales pitch, just an honest view of whether and how it can be fixed.
Book a 30-Min Process Review