SOL - Homeowners AI Chatbot

Portfolio, Ventures & What I Build

Product/systems design that turns AI tech into business results

I'm Gil, an AI Product Design Engineer. I take complex workflows and turn them into AI-native systems, from discovery through shipped product.

My north star is the complete practitioner: the idea that the best AI-native products are shaped by one mind that can hold design, engineering, and AI logic simultaneously, not handed off through layers of interpretation.

When the same person who defines the workflow also builds the prototype and wires the AI, translation loss disappears. Teams move faster and the gap between what we designed and what shipped closes.

What I Bring

Enterprise Systems

Enterprise End-to-End Systems Builder

I've designed complex product ecosystems at enterprise scale — multi-product platforms, multi-role workflows, and long-running transformation initiatives as a solo designer. Systems thinking, not just screen design.

AI Product Engineering

AI-Native Product Design Engineer

I design and prototype AI-native product systems — RAG pipelines, agentic workflows, human-in-the-loop interfaces, structured outputs, and context-aware products. AI product work grounded in real systems, not demos.

Education & Evangelism

Technical Educator & Evangelist

I help people understand, adopt, and build with new technologies — through demos, workshops, tutorials, and product systems thinking. Teaching as adoption; adoption as product design.

A consistent practice across technology cycles

Tech Speaker Microsoft · 5,000+ developers

Co-founder 8th Light · Digital transformation

Startups + Enterprise Engine Yard · ThoughtWorks · Dev tools

Founder Skowak · Design engineering · Teaching

Enterprise Focus SunPower · Digital transformation · Ecosystems

AI Eng Focus Craftal · CareDash · Multi Agent Systems

Now Contextus · Craftal v.2 · AI-native systems

How I Work

I don’t start with AI. I start with how work actually happens.

The method works in cycles. Each one scoped small enough to prove value fast, wide enough to compound across the whole system.

Each cycle makes the next one faster, safer, and more valuable. Real workflows before AI use cases. No baseline, no proof. No proof, no scale.

01 — 03

Understand

Discover
Model
Baseline

Know before you build

04 — 06

Intervene

Design
Adapt & Script
Measure

Build small, prove value

07 — 08

Promote

Promote
Orchestrate

Make it durable and connected

09 — 10

Scale

Govern
Reuse & Expand

Govern and compound

Where most AI projects break down

Most AI failures aren't technical. They're structural: siloed teams, AI built without real workflows, ideas that can't survive contact with production. The method closes those gaps.

Design, engineering, and AI in separate rooms

Work lands in three different hands. Every handoff is a translation, and translation is where ideas lose fidelity.

One practitioner across all three layers

I lead design, write code, and wire the AI as one strategic partner across all three disciplines. No handoffs, no interpretation layers, no gap between what was designed and what shipped.

AI built without a workflow to plug into

Teams rush to launch LLM features with minimal ROI. Without mapping real workflows or solving real problems, the result is unused chatbots and shallow automation.

Bridge between LLMs and real work

I map how work actually happens first, then wire the AI stack into those real decisions. The result is automation that runs, not a chatbot no one uses.

Technical Toolkit

The full stack, from workflow model to deployed AI system.

My stack spans the full system: AI engineering (OpenAI & Anthropic APIs, LangChain, LangGraph, DSPy, RAG pipelines, vector stores, structured outputs, agent orchestration), frontend and prototyping (React, Angular, Vue, TypeScript, API-connected interfaces), backend and tooling (Python, FastAPI, Rails, PostgreSQL/pgvector, Docker), and product and design (systems modeling, workflow mapping, information architecture, interaction design, design engineering).

As a product generalist, I choose the right tool for each layer (design, front end, data, or model) so ideas travel from workflow model to shipped system without unnecessary handoffs.

Getting personal

Things I do when I'm not building

Still-life sketching, coastal hiking, PC teardowns, and slow travel through cities with good architecture.

I draw to stay sharp. I travel to stay curious. I take things apart to stay honest about how they work.

Back

SOL - Homeowners AI Chatbot

An early chatbot experience that helped homeowners learn about solar, ask questions, and move toward savings estimates.

Skills applied:

Visual Design Interaction Design Conversational UI AI Strategy

Homeowners had many basic solar questions, but answers were scattered across website pages, sales calls, FAQs, and follow-up conversations.

I designed SOL as a guided conversational assistant that could explain products, answer common questions, collect intent, and connect homeowners to the next step.

The Problem

Solar education was too fragmented for curious homeowners.

Buying solar required homeowners to understand products, savings, roof suitability, installation, incentives, pricing, and long-term system value.

The website had information, but homeowners still had to search, read, compare, and decide what applied to their situation. If they had a specific question, they often needed to contact a sales rep.

That created unnecessary friction. Many visitors were not ready to talk to sales yet, but they still needed guidance before they could trust the process.

diagram sol 01 — Homeowners had to piece together answers across pages, calls, and follow-ups.

The challenge was to make solar education feel immediate, friendly, and guided without pretending the system could answer everything perfectly.

Solution

Creating a lightweight assistant inside the website.

I designed SOL to live directly inside the SunPower website experience. It started as a small assistant entry point that did not interrupt browsing, but stayed available when homeowners needed help.

sunpower sol 01 — SOL started as a subtle assistant entry point on SunPower’s website.

This mattered because the assistant needed to feel helpful, not intrusive. Homeowners could keep exploring the website while knowing support was one click away.

Making conversations feel persistent and approachable

The conversation drawer gave users a place to return to recent messages, assistant responses, and human updates.

sunpower sol 02 — Conversation history surfaced chatbot and human updates in one place.

This helped the experience feel less like a one-off widget and more like a lightweight communication layer across the solar journey.

The expanded panel gave the chat more room when users wanted to focus on a conversation.

sunpower sol 03 — The expanded panel gave conversations more room without leaving the page.

The design preserved context. Users could still see the website behind the panel, which kept the assistant connected to the broader product and education experience.

diagram sol 02 — SOL added a conversational layer on top of the existing website journey.

Guiding users with structured topics

Because this was pre-LLM chatbot technology, the experience needed structure. I designed topic shortcuts to help users start with common areas like Equinox, savings, pricing, and help.

sunpower sol 04 — Topic shortcuts helped homeowners start with common solar questions.

These shortcuts reduced the blank-page problem. Instead of forcing users to guess what the assistant understood, the interface suggested safe paths into the conversation.

This was important for trust. A constrained assistant that behaved predictably was better than an overly broad assistant that failed unexpectedly.

Turning answers into rich educational moments

SOL was designed to do more than return short text replies. It could surface structured cards, product explanations, images, and next-step prompts inside the chat.

sunpower sol 05 — SOL delivered product education directly inside the conversation.

The product card made the response feel tangible. Users could ask about SunPower panels and receive a concise explanation with supporting visual content.

This helped translate product knowledge into a more conversational, accessible format.

Moving from education to lead qualification

Once a homeowner showed interest, SOL could ask for the information needed to provide a savings estimate.

sunpower sol 06 — Address capture turned education into a savings estimate workflow.

The address form turned the chatbot from a passive support tool into a lead qualification path. It connected curiosity to action without forcing an immediate sales call.

Supporting continuity beyond the assistant

The experience also allowed human updates to appear in the same conversational space, including installation scheduling messages.

sunpower sol 07 — Human updates extended the conversation beyond the chatbot.

This was important because the solar journey does not end with a question. Homeowners need updates, reminders, and reassurance as they move from education into installation.

SOL hinted at a broader communication model: chatbot when possible, human support when needed, and one persistent place for updates.

Outcome

SOL made solar education feel more immediate and conversational.

SOL helped turn static solar education into a guided, interactive experience. Homeowners could ask questions, explore products, learn about savings, and move toward an estimate without immediately depending on a sales rep.

diagram sol 03 — The assistant turned questions into qualification and next-step intent.

Estimated impact based on project context:

Reduced friction for early solar education
Helped homeowners explore common questions faster
Created a softer path into lead qualification
Made product education feel more accessible
Reduced dependency on repetitive sales explanations
Established an early conversational UX pattern inside SunPower

The project was especially meaningful because it anticipated a direction that later became central to AI product design: conversational interfaces as guidance layers across complex product journeys.