Selected systems, not just deliverables.

The Challenge

iARx had no unified knowledge layer.

Critical product information lived across individual experts, shared drives, release threads, ticket histories, engineering conversations, and disconnected repositories. Software, hardware, operations, training, and support often described the same product differently. Content ownership was unclear, versioning was inconsistent, and employees had no reliable way to know whether the information they found was complete, current, or authoritative.

This was not only a documentation problem. It was an AI-readiness problem.

Generative AI cannot reliably answer questions when the knowledge beneath it is fragmented, duplicated, outdated, poorly structured, or missing context. Without a governed source of truth, any AI assistant would risk retrieving the wrong version, blending incompatible information, or producing confident but ungrounded answers.

Before iARx could scale AI, it needed a trusted knowledge foundation.

The Strategy

I approached the work as the design of an enterprise knowledge system — not simply the launch of a document library.

The goal was to create a governed information architecture that could serve people immediately while also supporting future AI-powered search, retrieval-augmented generation, workflow automation, and intelligent content experiences.

The operating model connected four layers:

What I Led

Built the company’s first enterprise source of truth

I created the documentation hub from zero, bringing together knowledge across software, hardware, operations, support, implementation, and third-party products.

The hub established a shared information architecture across previously disconnected domains, giving teams a clear place to publish, discover, maintain, and govern trusted product knowledge.

Rather than treating documentation as a collection of files, I designed the hub as a connected knowledge layer.

Designed an AI-ready content architecture

I structured the repository to support both human navigation and machine retrieval. This included:

• Product and domain taxonomies
• Version-aware content structures
• Standardized document types
• Audience and distribution controls
• Named content owners and subject-matter experts
• Review and maintenance signals
• Consistent metadata across software and hardware
• Clear separation of customer-facing and internal knowledge
• Reusable content patterns that improved retrieval precision

These structures created the semantic context AI systems need to retrieve the right content instead of merely locating documents that contain similar words.

The result was a stronger foundation for semantic search, enterprise copilots, vector retrieval, and RAG-based question answering.

Integrated AI into the knowledge experience

AI was not added as a surface-level chatbot after the hub was built. It was incorporated into the broader knowledge strategy.

I evaluated how employees could use generative AI tools to search, summarize, interpret, and apply product knowledge while remaining grounded in approved enterprise content. The work included:

• Preparing high-value documentation for AI-assisted retrieval
• Identifying authoritative source content for grounding
• Improving content structure for chunking and semantic indexing
• Reducing duplication and contradictory source material
• Establishing metadata that could support retrieval filtering
• Exploring retrieval-augmented generation across the documentation corpus
• Designing use cases for copilots and internal knowledge assistants
• Testing prompt patterns against real support and product questions
• Identifying hallucination risks, missing context, and retrieval failure modes
• Evaluating answer quality based on accuracy, completeness, source alignment, and usability

This shifted the conversation from “How do we add AI?” to “What knowledge does AI need in order to be trustworthy?”

Improved retrieval quality and grounding

One of the core risks in enterprise AI is not model capability — it is weak source material.

I strengthened the knowledge base so AI-generated responses could be grounded in content that was:

• Authoritative
• Current
• Version-specific
• Audience-appropriate
• Traceable to an owner
• Structured consistently
• Connected to the correct product context

This work helped reduce the conditions that lead to hallucinations, outdated answers, context collapse, and cross-version confusion. It also created the foundation for source-cited responses, human-in-the-loop review, retrieval evaluation, and confidence-based escalation.

Established a biweekly change-intelligence system

I developed a repeatable release-communication process that translated product and engineering changes into clear operational knowledge. The system connected:

• Product decisions
• Software releases
• Hardware changes
• Support readiness
• Training needs
• Documentation updates
• Customer-facing communication

Instead of allowing change information to remain buried in engineering threads or isolated release artifacts, the process surfaced what changed, who was affected, what action was required, and which knowledge assets needed to be updated.

This created a more reliable information supply chain for both employees and future AI systems.

Built cross-functional knowledge governance

I established an operating model across Product, Engineering, Hardware, Support, Training, Field Service, Solutions, and customer-facing teams. The governance model clarified:

• Who owned each knowledge domain
• Who approved technical accuracy
• When content should be reviewed
• How version changes were handled
• Which content was authoritative
• What could be exposed to customers
• What could be used to ground AI
• When human review was required

This transformed documentation from an isolated writing function into a shared enterprise capability.

AI Architecture and Design Principles

The knowledge system was developed around several principles essential to trustworthy enterprise AI.

The Outcome

The documentation hub showed sustained quarterly growth and became the foundation for how teams found and shared trusted product knowledge.

Support, training, product, engineering, field service, implementation, and customer-facing teams gained a clearer path to accurate information. The biweekly change system improved visibility into product updates, while the governance model reduced ambiguity around ownership, review, and authority.

The broader impact extended beyond documentation. The organization gained:

• A governed enterprise knowledge layer
• Improved readiness for AI-powered search and assistance
• Stronger grounding data for RAG and copilots
• Better visibility into product and operational change
• Reduced repeated clarification work
• Clearer content ownership and lifecycle management
• A more scalable foundation for support automation
• Lower risk of AI retrieving outdated or contradictory information
• A repeatable model for connecting knowledge, workflows, and intelligent systems

What began as a documentation initiative became part of the company’s AI infrastructure.

Why It Mattered

Many organizations begin their AI strategy with a model, chatbot, or platform. This work began one layer deeper.

It created the governed knowledge environment those technologies depend on: trusted sources, structured context, clear ownership, retrieval signals, version control, and operational feedback loops.

The result was not simply a better place to store documents.

It was a system that helped people find what they needed, trust what they found, apply it confidently, and extend that knowledge safely into AI-enabled workflows.

Capabilities Demonstrated

The Challenge

Coinbase Institutional supported organizations and individuals moving significant assets through complex, rapidly evolving financial products.

The audience included institutional investors, ultra-high-net-worth individuals, and major financial entities, including BlackRock. These customers needed information they could act on confidently — whether they were navigating account access, custody, transactions, security controls, or operational requirements.

The content carried an unusually high burden of precision. A single unclear instruction, outdated detail, or inconsistent explanation could create financial, regulatory, security, or reputational risk.

At the same time, the business was moving quickly. Product functionality changed, policies evolved, and support teams needed current guidance without allowing speed to weaken editorial control.

What I Led

Established editorial rigor for institutional content

I led editorial quality and review discipline across institutional support content, creating a consistent standard for accuracy, clarity, structure, tone, and customer usability.

I translated highly technical and operational information into plain language without oversimplifying the underlying financial or security implications.

The goal was not merely to make content readable. It was to make it dependable under pressure.

Built a four-function review model

I developed more predictable review workflows across:

• Legal
• Compliance
• Product
• Support

Each function had a distinct role in validating regulatory language, product accuracy, operational practicality, and customer clarity.

This replaced fragmented, ad hoc feedback with a more structured editorial process and helped surface risk earlier — before content reached institutional clients or frontline support teams.

Created consistency across complex workflows

Institutional content often crossed multiple systems, teams, products, and stages of the customer journey.

I standardized terminology, instructional patterns, escalation language, and editorial conventions so customers received a coherent experience — even when the underlying information came from different subject-matter experts.

This reduced ambiguity and made complex workflows easier to follow, verify, and maintain.

Balanced speed with risk management

The challenge was not choosing between precision and speed. It was designing a system that could support both.

I introduced editorial checkpoints, reusable standards, and clearer approval paths so content could move efficiently while preserving the level of scrutiny expected for high-value financial relationships.

Editorial review became part of the operating model rather than a final-stage bottleneck.

Strengthened the knowledge foundation

This work took place during the early expansion of enterprise AI and intelligent support systems.

While the primary focus remained editorial quality, the same disciplines — structured content, consistent terminology, authoritative sourcing, clear ownership, and controlled review — also created stronger foundations for search, automation, machine-assisted support, and future AI-enabled knowledge retrieval.

The principle was simple: intelligent systems are only as trustworthy as the content they rely on.

The Outcome

Institutional clients received clearer, more consistent support content for complex and high-risk financial workflows.

The broader content operation gained:

• More predictable review cycles across four critical functions
• Stronger consistency across institutional guidance
• Clearer ownership and approval paths
• Earlier identification of legal, compliance, and operational risk
• Faster translation of product change into customer-ready content
• A more structured knowledge base for search, support, and future automation

Most importantly, editorial rigor became embedded in how the team shipped.

It was no longer a final review step. It became part of the infrastructure supporting trust with some of the company’s most sophisticated and consequential customers.

Capabilities Demonstrated