An AI-powered tax strategy agent
for high-net-worth clients
High-net-worth clients face tax situations too complex for generic software and too fast-moving for once-a-year advisor meetings. This prototype shows how an AI agent can fill that gap — instantly matching a client's specific life and financial situation, depicted in a prior year's tax return, against a curated playbook of 111 proven tax strategies to surface recommendations that would otherwise take hours to produce. This product is envisioned as an in-house productivity tool for the tax advisor in a Virtual Family Office (VFO). Sensitive client data stays within the VFO. For more details, see the Product Discovery tab. In this product, the Master Tax Strategy Playbook will be proprietary to the VFO and thus will be its competitive advantage.
Built on n8n, Pinecone, and OpenAI. Client tax return data is retrieved from Google Drive, embedded using text-embedding-3-small, and matched against a 111-strategy vector database via semantic search. An LLM-based intent router (gpt-4o-mini) classifies each query and routes it to the appropriate analysis pipeline, returning advisor-grade recommendations in a conversational interface. Five simulated HNW client personas are pre-loaded for the demo. For more details, see the Architecture tab.
Who is this actually for? Why does it exist?
My primary goal was to get hands-on with AI agent tools and build a prototype I can showcase to hiring managers and recruiters — something that addresses a real business opportunity and requires genuine product thinking. I also wanted to demonstrate I could apply product sense to a domain far from my core expertise: consumer devices, displays, and color imaging.
This prototype supplements my UW Foster MBA elective from Winter '26 — Analytics Consulting Lab — for a consulting project titled A Business Case for an AI-Powered CPA Practice. The Market Analysis tab draws directly from that work. Our "client" was a Chicago-based wealth management firm serving High-Net-Worth (HNW) clients, with interest in acquiring and scaling accounting practices. My conversations with their managing director and accountant were essentially product discovery interviews.
Key insights from discovery:
- HNW individuals don't feel their own tax pain — they delegate complexity entirely. That's why they have CPAs.
- Clients are unlikely to act on AI-generated tax strategy without CPA validation.
- Delivering tax advice to end consumers without a licensed practitioner raises unauthorized practice concerns.
- Even with tax returns stored server-side, a client-facing flow creates consent and liability issues that a B2B internal tool avoids entirely.
To sidestep data privacy concerns around uploading real tax information, I built five HNW personas with varied life and tax situations. That said, I plumbed the architecture in a way that enabling personalized recommendations for real users would be straightforward.
The current chat interface already generates outputs in a few clicks. In a real product, a key feature would be generating a PDF report with personalized recommendations for CPA review and client sharing — out of scope for this prototype. I see this as a proof-of-concept; the next step, if I continue, would be an MVP (minimum-viable-product).
My secondary goal is testing product-market fit for these two hypotheses.
I'm looking for data-driven signals — so I'd be genuinely grateful if you took the 30-second, 3-question survey. Not specific to HNW individuals; anonymous submissions welcome.
Three-Workflow Pipeline
Each workflow handles a distinct stage — from ingesting tax return data to retrieving strategies to generating advisor-grade output.
Five HNW Personas
Each persona represents a distinct high-net-worth archetype with simulated TY2025 tax return data.
User Flow
From workflow design to real-time strategy — a walkthrough of the prototype in three steps.
