Well-designed knowledge
is the fuel for GenAI.
Most AI projects stall not because the models are bad, but because the foundation is. I design knowledge architectures, voice agents, and AI experiences engineered to perform from the first real user, not just the demo.
Four things I do exceptionally well
Not a consultancy. Not an agency. I work directly with founders and product teams on the specific problems where deep AI expertise makes the biggest difference.
Knowledge Base Design
The difference between an AI that answers correctly and one that confidently hallucinates is almost always the knowledge layer. I design retrieval architectures, content taxonomies, and chunking strategies that give your AI the right information at the right time.
- Retrieval-augmented generation (RAG) architecture
- Content taxonomy and metadata strategy
- Chunking, embedding, and indexing optimization
- Knowledge quality audits and gap analysis
Voice Agents
Phone and web-based voice agents that handle real conversations, not just the easy ones. From intent design to fallback flows, I build agents that hold up under the pressure of actual users with actual problems.
- Conversational flow and intent architecture
- Real-time speech-to-text and TTS integration
- Fallback handling and escalation logic
- Ongoing tuning based on conversation data
Synthetic Video Avatars
Cloned and AI-generated video presenters for training content, onboarding, customer communications, and more. Scale your message without scaling your production costs.
- Voice and likeness cloning for consistent brand presence
- Script-to-video pipelines for rapid content production
- Multilingual localization without reshooting
- Integration with LMS and content delivery platforms
AI Coaching Platforms
End-to-end platforms that combine personalized AI with structured learning paths. Built for engagement and measurable behavior change, not just completion rates.
- Adaptive learning path architecture
- AI coach persona design and prompt engineering
- Progress tracking and personalization logic
- Integration with existing HR and LMS systems
Built by a product person who actually builds
I came up through product management, which means I think about user needs, delivery constraints, and what "done" actually looks like. AI is the tool, not the point.
Foundation first
Every AI project I take on starts with the knowledge layer. What does the system need to know? Where does that content live? How fresh does it need to be? Getting this right before writing a single line of agent code is what separates working systems from expensive prototypes.
Ship, then iterate
I have a product management background. That means I think in terms of working software, not decks. I scope things so they can go live, then improve based on what real usage reveals. You get momentum and you get learning, not a 6-month design phase.
Build for the edge cases
Happy-path demos are easy. The hard work is designing for the user who asks the question you didn't expect, the caller who rambles, the learner who falls behind. I spend disproportionate time on the scenarios that break things.
Your AI is only as good as what it knows.
Language models are powerful, but they are also generalists. What makes them useful inside your product is the specific, well-structured knowledge you give them access to. That means getting the information architecture right before you write the first prompt.
I have spent years working on exactly this problem. Not as a researcher, but as a practitioner who has designed knowledge systems for production AI products. The work is part content strategy, part information architecture, and part systems thinking. When it is done well, your AI stops being a liability and starts being a genuine asset.
Let's figure out if I can help
Tell me what you are working on and where you are stuck. I will let you know honestly whether it is something I can add real value to.