Chris Fornesa Project Guide

The Project Guide Agent

Who is this for?

This is for any visitor interested in learning more about my skills, projects, past work, and professional background, presented in an AI-generated web page. For example, users can prompt for specific queries, such as their educational background, experience, expertise, or projects tailored to something they're interested in. From there, the agent will generate a web page with content that's suited to the user's query. Unlike a static portfolio website, the user can learn more about me and evaluate whether I am a good professional fit.

General Inquiry User Personas

Persona 1. The "Cold Start" Discovery (Unfamiliar User)

The Scenario: A visitor lands on the page through a shared link and has no idea who you are or why this agent exists.

The Query: "I just landed on this page, and I'm not sure what I'm looking at. Who is Christopher Fornesa, and what are the three most important things I should know about his professional expertise? Give me a few options for what I should explore first."

Synthesis – Can it summarize my transition from IT to Data Science into a simple, jargon-free value proposition?

Persona 2. The "Credibility Check" (Unfamiliar User)

The Scenario: A potential collaborator sees the "4.0 GPA" and "Data Science" claims but wants to see if there is actual professional weight behind the academic stats.

The Query: "Christopher has impressive academic stats, but does he have real-world experience? Explain his professional background and show me how his previous work in industry informs the way he builds technical projects today."

Cross-Registry Synthesis – Can it link the "Operational Maturity" (Static Knowledge) to my professional history without being explicitly told "Look at the Chevron data"?

Persona 3. The "Innovation Scout" (Semi-Familiar User)

The Scenario: Someone is specifically interested in AI/Automation but doesn't know your specific project names (like Tanaga).

The Query: "I'm interested in AI Agents and LLMs. What is Christopher's philosophy on 'Agentic AI,' and can you show me a directory of the autonomous systems he has actually built?"

Registry Isolation – Does it stick to the Agentic portfolio while omitting the Political Polarization Capstone unless it's relevant to the "Agent" methodology?

Persona 4. The "Technical Auditor" (Informed)

The Scenario: A Senior Engineer or Lead Data Scientist is reviewing your work and wants to see "under the hood" to verify your engineering discipline.

The Query: "Explain how Christopher managed high-compute resource constraints in his machine learning work. I want to see a specific example of his engineering strategy for preventing memory overflows, including a Python code snippet."

Static Reliability – Does it use the "Modular Notebook Separation" facts to provide a substantive answer, even if it can't crawl the live code in that moment?

Persona 5. The "Recruiter Match" (Professional Intent)

The Scenario: A recruiter has a specific job description for a "Machine Learning Engineer" and wants a quick "Fit Analysis" without having to read your entire site.

The Query: "I'm hiring for a role that requires a mix of Enterprise IT Systems and Advanced NLP (Transformers). Analyze Christopher's fit for this role based on his history and provide buttons so I can contact him immediately if he's a match."

Contact Hub Integration – Does the final output properly feature the Bento-style buttons for LinkedIn, Email, and Phone?

Capstone Inquiry User Personas

Persona 1. The "Metric-Obsessed" Hiring Manager (Targeted Inquiry)

The Scenario: A Data Science Manager at a tech firm is filtering candidates. They don't have time for fluff; they want to know if you understand model evaluation beyond simple "Accuracy." They are specifically looking for competence in handling Macro F1 scores in imbalanced datasets.
The Query: "I'm looking at your Political Polarization Capstone. Can you walk me through the specific performance results for the Convabuse and MLMA datasets? I’m particularly interested in the Macro F1 scores and which model architecture—Logistic Regression or Neural Networks—yielded the best results for these specific high-risk content categories."
Dynamic Retrieval Check:

• Does it correctly identify the specific target metrics?

• Does it distinguish that while Neural Networks were "best" for minority isolation, Weighted Logistic Regression was statistically competitive?

Persona 2. The "Baseline Skeptic" (Risk Assessor)

The Scenario: A Senior Lead Data Scientist is skeptical of "perfect" portfolios. They know that real-world data is messy. They want to see if you are honest about what didn't work, as this reveals your understanding of the data's difficulty.
The Query: "Looking at your Political Polarization Capstone, which models performed the worst on the Convabuse and MLMA datasets? I want to understand the baseline failures before looking at the optimized results."
Truth-Anchoring Check:

• Does it bravely display the 0.40 – 0.55 F1 range for the Linear/Imbalanced baselines?

• Does it explain why they failed (e.g., "collapsed under high dimensionality" or "majority class overfit") rather than just hiding the bad numbers?

Persona 3. The "Executive" Stakeholder (Big Picture)

The Scenario: A non-technical stakeholder (Product Manager or VP) lands on your site. They don't care about "F1 scores" or "Hyperparameters." They want to know the business value and the scope of what you built.
The Query: "Can you give me a high-level summary of your Political Polarization Capstone? What were the main datasets and what did you achieve?"
Narrative Synthesis Check:

• Does it abstract away the math to focus on the "40% Lift" via SBERT (a type of neural network model)?

• Does it clearly list the four datasets (Convabuse, MLMA, DGHS, US2020) without getting bogged down in the weeds?

• Does it sell the "Professional Context" (Chevron experience + Academic rigor)?

Persona 4. The "Code Auditor" (Technical Validation)

The Scenario: You have passed the initial screen. Now, a peer engineer is digging into your methodology to ensure you didn't just copy-paste code. They want to verify your implementation of advanced techniques like SMOTE and Cost-Sensitive Learning.
The Query: "I'm doing a technical audit of your Political Polarization Capstone. Walk me through how you handled class imbalance across the different datasets—what specific techniques did you implement?"
Methodology Mapping Check:

• Crucial: Does it link specific techniques to specific notebooks? (e.g., linking "SMOTE" to Sentiment_Analysis_Splits_SMOTE.ipynb).

• Does it explicitly mention the "Operational Maturity" of using modular batching and weighted loss functions?

Methodology: The Intent-Based Discovery Engine

The Chris Fornesa Project Guide utilizes a Strategic Intent-Routing Protocol that goes beyond static data retrieval. It is engineered to identify the user's specific "onboarding needs"—whether they are looking for a high-level biography or a deep technical audit of your enterprise history.

• Intent-Based Routing: The agent analyzes user queries to distinguish between [INTENT: GUIDED_TOUR] (new visitors) and [INTENT: TECH_AUDIT] (skeptical technical reviewers), ensuring responses are calibrated to their level of familiarity.

• Cross-Registry Synthesis: By reasoning across separate data "Registries" (Career, Project, and AI Focus), the system connects disparate milestones—like my 5-year tenure at Chevron—to your current high-performance Data Science results at Boston University.

• Operational Contextualization: The methodology ensures that technical findings, such as SBERT gradient mapping (+40% accuracy) or modular notebook separation, are presented as evidence of professional engineering maturity rather than just academic theory.

Technical Architecture: The Multi-Registry Retrieval Protocol

An agentic portfolio advocate requires a resilient infrastructure capable of handling complex, unstructured inquiries in real time. This system implements a Modular Knowledge-Retrieval Architecture to ensure accuracy and professional alignment.

The Solution: This agent implements Data Integrity Sovereignty via a Triple-Registry Protocol. Every consultation is powered by a dynamic search across three isolated knowledge bases, ensuring high-fidelity results even for visitors with zero prior context:

1. Direct Career Registry: Pulls from my interactive resume to verify academic excellence (4.0 GPA) and specific professional milestones.

2. Direct Project Registry: Extracts technical methodologies and results (e.g., Random Forest vs. Transformer benchmarks) for deep-dive technical audits.

3. AI Focus Registry: Dedicated to emerging tech, specifically showcasing my vision for Agentic AI and LLM orchestration.

Static Fallback Logic: To prevent "hallucinations" or broken experiences, the architecture includes a Zero-Placeholder Guardrail. If live crawls are unavailable, the system automatically defaults to a core set of immutable "Static Knowledge" facts, such as my BU MS and Chevron IT Analyst background.

Prompt Guide for High-Value Results

To generate the most accurate and actionable insights from the Project Guide, use prompts that focus on comparative career analysis and technical methodology:

• The "Fit Analysis" Prompt: "Analyze Christopher's background for a Senior ML Engineer role. Map his 5-year tenure at Chevron and his MS in Data Science findings directly to the requirements of high-compute infrastructure and NLP optimization."

• The "Methodology Deep-Dive" Prompt: "Explain the specific technical problem Christopher solved regarding memory overflows during model training. Provide the architectural reasoning for his modular notebook separation strategy."

• The "Guided Discovery" Prompt: "I am unfamiliar with Christopher's work. Provide a high-level overview of his professional evolution and suggest three distinct technical paths I should explore to understand his expertise."

Explainer

Technical Insight: This interactive agent demonstrates the transition from raw research to a production-ready Agentic Portfolio System. While the early phase of the project focused on fine-tuning SBERT and Random Forest models for political polarization data, this live implementation showcases autonomous Intent-Based Routing. The system is capable of real-time synthesis across three isolated registries (Career, Project, and AI Focus), ensuring that professional advocacy is both data-driven and contextually aware.