AI Portfolio Assistant
Building a trustworthy AI assistant wasn't primarily a cloud project—it was a product strategy project.
The technical implementation of my portfolio assistant took only a few days. The work that made it successful took months.
Before connecting a single AI model, I spent months refining my professional identity, documenting my experience, clarifying my product philosophy, writing long-form perspectives, defining truthful positioning, identifying the boundaries of my expertise, and deciding how I wanted to represent my career. The AI assistant became the interface to that knowledge rather than the source of it.
This project demonstrates my approach to AI product management: successful AI systems begin with trusted knowledge, thoughtful information architecture, clear guardrails, and an understanding of user needs. The model itself is only one component of the solution.
You can access my AI Portfolio Assistant via the chat bubble on any page, or the full-page assistant is here!
AI Product Design
1. Knowledge Architecture & Retrieval
Rather than relying solely on the language model's general knowledge, the assistant is grounded in a curated Retrieval-Augmented Generation (RAG) architecture. The knowledge base combines structured portfolio content, project documentation, resumes, published perspectives, and carefully defined system instructions to ensure responses remain accurate, consistent, and grounded in trusted information. This approach reflects my belief that successful AI products begin with trusted knowledge rather than simply larger models.
2. User Experience & Conversation Design
The experience was intentionally designed to help visitors quickly understand my background without requiring them to navigate dozens of pages. Welcome messaging, suggested questions, conversation flow, and progressive disclosure guide users toward the topics most relevant to recruiters, hiring managers, and professional peers while allowing them to explore additional detail naturally through conversation.
3. Trust, Guardrails & Product Governance
The assistant was designed with clear behavioral boundaries rather than simply maximizing conversational ability. System instructions establish truthful positioning, define the limits of my documented experience, provide canonical responses for high-value questions, and specify how the assistant should respond when information is unavailable or uncertain. The objective is to create a trustworthy product that consistently represents my experience without exaggeration or hallucination.
4. Product Thinking Through Hands-On Prototyping
Although this project uses Google Cloud infrastructure, the primary objective was never simply to build a chatbot. It was to explore the complete lifecycle of an AI product—from defining the user problem and organizing trusted knowledge to designing conversation flows, evaluating outputs, and iteratively improving the experience through structured testing. Working directly with Vertex AI, Google Cloud Storage, Data Stores, and Agent Builder provided valuable hands-on experience while reinforcing the importance of balancing product strategy, technical feasibility, and responsible AI design.
5. Evaluation & Continuous Improvement
The assistant was developed using an iterative evaluation process rather than a one-time implementation. Structured acceptance testing, representative recruiter and hiring manager scenarios, canonical responses, and ongoing refinement were used to improve factual accuracy, conversation quality, and job-fit reasoning. This reflects the same product philosophy I apply to enterprise products: successful AI systems improve through thoughtful evaluation, user feedback, and continuous iteration rather than model changes alone.
Technologies Used
Google Cloud Console
Google Cloud Storage
Vertex AI Agent Builder
Gemini
Dialogflow Messenger
Retrieval-Augmented Generation (RAG)
Prompt Engineering
Knowledge Architecture
Conversation Design
Information Governance
Acceptance Testing
Human-Centered Product Design
The Problem
Recruiters and hiring managers often have only a few minutes to understand a complex twenty-year career. A traditional resume cannot fully explain product philosophy, healthcare experience, leadership style, project context, or the reasoning behind career decisions. I wanted to create an assistant that could answer those questions conversationally while remaining truthful, transparent, and grounded in approved information.
Product Goals
Help recruiters quickly determine mutual fit.
Answer common questions consistently.
Explain the broader scope of my experience beyond a formal job title.
Evaluate job descriptions against my documented experience and career goals.
Guide visitors toward relevant portfolio content rather than replace it.
Maintain strict boundaries around truthfulness and confidentiality.
Knowledge Architecture
The assistant is grounded in:
a structured professional knowledge base,
published portfolio content,
resumes,
selected project documentation,
and carefully defined system instructions.
The knowledge base separates public facts from private behavioral rules, defines truthful positioning, documents the limits of my expertise, establishes canonical responses for high-value questions, and includes an acceptance test plan used to validate the assistant before publication.
This was the most time-consuming part of the project and, ultimately, the most important.
More Than a Chatbot
Building this portfolio assistant reinforced a principle that has shaped my approach to product development throughout my career: artificial intelligence is most valuable when it amplifies thoughtful human work rather than attempting to replace it.
Although the technical implementation of the assistant took only a few days, the knowledge architecture behind it evolved over months. Before connecting a language model to my portfolio, I had already spent considerable time clarifying my professional identity, documenting projects, defining product philosophies, writing long-form perspectives, identifying the boundaries of my experience, and carefully deciding how I wanted to represent my career. AI accelerated the organization, retrieval, and presentation of that knowledge, but it could not determine what was important, what was true, or how those experiences should be interpreted.
That distinction closely reflects my broader philosophy toward AI-enabled healthcare products. Successful AI systems are built on trusted information, clear business definitions, thoughtful workflows, and disciplined product thinking. The quality of the experience depends far more on the quality of the underlying knowledge than on the sophistication of the model itself.
Ultimately, this project demonstrated the same principle I have seen repeatedly in healthcare product development: technology creates the greatest value when it helps people spend less time organizing information and more time applying judgment, experience, and expertise where they matter most.