I’m Chris, a CTO who connects product thinking, system design, and hands-on development to build applications with RAG and agentic AI—grounded in knowledge, orchestrated to get useful work done.
That means connecting what’s documented with what people know—and capturing the expertise that hasn’t been written down yet.
FROM THE FIRST QUESTION TO THE WORKING PRODUCT.
RAG workflow: existing documentation and unwritten expertise captured as notes, audio, or video become the sources. These sources are extracted, transcribed, chunked, and embedded into a scoped knowledge index. At query time, agentic orchestration can plan queries, coordinate retrieval tools, and refine weak evidence. Relevant evidence is retrieved and reranked, then supplied with the question to a language model. The model generates an answer with source citations. This diagram shows knowledge grounding and coordination, not permission to change hardware.
Existing documentation + recorded expertise
Embeddings + a scoped vector index
Relevant evidence
The LLM composes from the evidence.
With citations back to the sources.
Plan queries · coordinate tools · refine retrieval
Capture knowledge. Ground the answer.
Orchestrate the work—with people in control.
01 / SELECTED WORK
A knowledge engine, an event-and-media platform, and a hardware assistant. RAG supplies the context; AI workflows and agentic orchestration connect it to useful work—with people in control.
01 THE KNOWLEDGE INFRASTRUCTURE
One platform. Every kind of knowledge.
A multi-tenant RAG platform that makes existing documentation and newly captured expertise searchable through documents, audio, and video—with answers grounded in their sources.
02 THE EVENT-TO-KNOWLEDGE EXPERIENCE
The event ends. The knowledge keeps working.
A RAG-connected event and video platform that helps audiences discover what was said on stage—and gives editorial teams control over how that knowledge is published.
Illustration: equipment manuals and best practices feed a RAG-powered assistant’s configuration recommendation. A separate user approval is required before settings on the video processor are changed. The illustrated rack-mounted processor is generic and this portfolio is not connected to hardware.
No approval. No hardware changes.
03 THE APPROVAL-GATED ACTION LAYER
From technical knowledge to approved action.
An assistant connected to the RAG platform that uses equipment documentation and best practices to guide a natural-language conversation, recommend video-processor configurations, and change hardware settings only after user approval.
Capture knowledge. Ground decisions. Orchestrate work. The RAG platform combines source-grounded retrieval with AI query planning. SDG Media Zone turns media knowledge into event discovery, semantic moments, and AI-assisted editorial work. The hardware solution connects recommendations to action—but only with user approval.
Explore Media Zone02 / MY APPROACH
I’m interested in the full journey—not just the model. A good application begins with the right problem and ends with something people can depend on.
The technology is the means.
The working product is the point.
Start with the operational need. Map the people, existing sources, and the expertise that hasn’t been documented. Plan how to capture that know-how alongside the constraints and decisions the application should support.
Connect the user journey to RAG and agentic orchestration. Define the tools, evidence quality, access boundaries, and approval points—and choose where model-assisted planning or a deterministic workflow is the right fit.
Initiate the delivery, develop the application end to end, and account for the things that keep it usable: security, background work, monitoring, and deployment.
BEYOND THE DOCUMENTATION
Existing documentation is a powerful starting point. But the practical details often live with one person: how to handle an exception, why a setting matters, or what to do when the standard procedure falls short.
This is tribal knowledge, or tacit expertise. Capturing it turns unpublished experience into a source the application can use—not something the next person has to rediscover.
Don’t just search what’s written.
Capture what’s known.
Record an expert interview, narrate a hands-on walkthrough, or write down the steps and workarounds that aren’t in the manual.
INTERVIEWS · WALKTHROUGHS · NOTESDocument why a decision works, when it doesn’t, and which exceptions matter. Review the captured material with the contributor before relying on it.
DECISIONS · EXCEPTIONS · EXPERIENCEBring those notes and recordings into the platform alongside existing documentation. Transcription and retrieval make the expertise available to the application—and the wider team.
CAPTURED KNOWLEDGE → USEFUL APPLICATIONSThe goal is richer, relevant knowledge—not simply more data. It must be captured first; RAG cannot retrieve what has never been recorded.
A FIELD NOTE ON HOW IT WORKS
RAG connects AI to a defined collection of knowledge. Agentic orchestration builds on that grounding to plan queries, coordinate tools, and refine the evidence used in a response.
THE ORCHESTRATION LAYER
Retrieval supplies the relevant knowledge. Agentic orchestration uses model-assisted planning and tools to coordinate the steps needed to reach a useful result—not just produce a reply.
Break a question into focused queries—or ask a follow-up when the request is ambiguous.
Search the relevant knowledge sources, combine the results, and rerank the evidence.
When retrieval is weak, use a repair pass before composing a source-grounded response.
Orchestration is not blanket permission. The hardware assistant can recommend a configuration, but changes settings only after user approval. In SDG Media Zone, AI helps prepare content; editors decide what becomes public by adding videos to events.
Agentic where planning and adaptation help. Deterministic where a predictable workflow is better. Human approval where hardware changes are involved.
Existing documents + captured notes, interviews, and walkthroughs
Chunk text, transcribe media, add visual context
Scoped text and video vector indexes
Source evidence, ready for retrieval
Normalize, clarify, and plan queries
Find, deduplicate, refine, and repair
Stream a response with citations and clips
Web workspaces · SDG Media Zone · Hardware integrations
The RAG platform’s optional LLM query planner can generate 1–4 sub-queries or ask for clarification instead of searching. Its orchestrator fans queries out with bounded concurrency, reranks results, and can make a repair pass when evidence is weak. This is the adaptive part of the workflow; not every background worker or scheduled AI job is an autonomous agent.
Transcript indexing can happen before asynchronous visual enrichment updates the video records. Audio is searchable through transcription—not a separate audio vector index. Retrieval stays within a tenant’s configured workspace, and citations let people inspect the evidence. These are documented design choices, not a guarantee that AI outputs are always correct.
03 / WHAT I BRING
Strategy, architecture, experience, and engineering are not separate handoffs. They’re parts of the same product.
Problem framing, workflow design, knowledge organization, and capturing the unwritten expertise an application needs to be useful.
Multimodal ingestion and source-grounded retrieval, combined with model-assisted planning, tool orchestration, and adaptive retrieval within defined boundaries.
Thoughtful interfaces, typed APIs, reusable SDKs, public experiences, and hardware integrations that apply changes only after user approval.
Tenant isolation, delegated access, background workers, privacy-conscious analytics, and deployment design.
TOOLS IN THE WORK
THE ORGANIZATIONAL VALUE
Capture unwritten procedures, practical workarounds, and decision context. Make that know-how searchable alongside existing documentation so the wider team—and its applications—can use it.
Coordinate multi-step AI work instead of stopping at a chat response. Keep decisions grounded in evidence, preserve editorial control, and require user approval before changing hardware settings.
Connect a reusable RAG foundation to internal workspaces, event-and-media discovery, and approval-gated hardware operations.
These are the workflows the platforms enable—not unverified claims of time saved, revenue generated, or measured business impact.
LET’S BUILD SOMETHING USEFUL
I connect product direction, technical architecture, and hands-on development to turn organizational knowledge into useful AI applications. Let’s talk about what you’re building.
A real conversation. No pitch deck required.