The Problem I’m Solving

Have you ever Googled “AC not cooling” at 11 PM, landed on a 40-minute video, and realized halfway through that the equipment on screen looks nothing like yours?

That’s the current state of DIY repair guidance: scattered, generic, and often unsafe.

Most people face one of two broken choices:

  1. Follow fragmented online advice — hoping random forum posts or YouTube videos match their specific situation
  2. Call a technician immediately — because they have no reliable way to assess whether this is actually a DIY problem

There’s a huge gap in the middle. A space where you could safely explore whether something needs a professional, with structured guidance that knows when to stop and escalate.

That’s why I’m building XpertConnect.

XpertConnect is an AI-first DIY guidance platform designed to help homeowners troubleshoot problems safely and systematically — and then hand off to a real expert when the situation requires it.

This post is a public build update: where the project stands, how I’m approaching the architecture, and why I believe the answer isn’t “make an AI chatbot” but rather “build structured workflows that AI supports.”

I’m keeping this intentionally public-safe. That means sharing the vision, philosophy, and architectural direction — but not exposing sensitive implementation details or decisions that belong in private documentation.

The Vision

Here’s the simplest way to describe XpertConnect:

Structured troubleshooting with AI guidance, plus a clear path to expert escalation when DIY ends.

The platform has four core pillars:

AI DIY Copilot — Step-by-step troubleshooting flows that guide, not improvise. Users move through a progression of checks, not an open-ended conversation.

Safety-First Boundaries — Explicit guardrails around what a homeowner should and shouldn’t attempt. Not to prevent people from trying, but to protect them when they do.

Expert Escalation — A clear handoff mechanism for problems that are unsafe, specialized, or inefficient to keep troubleshooting alone.

Future Marketplace Foundation — Infrastructure for connecting users with vetted service providers when escalation happens.

The key insight is that safety isn’t a feature — it’s a product requirement. Everything else serves that foundation.

Safety Comes First

If XpertConnect™️ fails at one thing, it can’t be safety.

That means dangerous work categories are treated fundamentally differently from safe ones. Not to discourage people from trying — but to ensure that when they do, they’re making informed decisions with appropriate warnings.

The safety model operates in three tiers:

Low-risk tasks can be guided clearly and systematically. Think: filter changes, thermostat adjustments, cleaning lint traps.

Moderate-risk tasks require stronger warnings and contextual checks. Users must acknowledge the risk before proceeding. Think: draining systems, checking refrigerant lines, electrical diagnostics.

Professional-only work triggers escalation instead of more instructions. These are tasks where DIY attempts create liability, damage, or safety hazards. Think: electrical repairs, major component replacement, certified inspections.

The core principle is honest: a trustworthy product knows when to stop giving guidance and start redirecting.

That boundary is non-negotiable.

Starting With HVAC

The first focus area is HVAC, specifically one of the most common problems homeowners face:

“Our AC isn’t cooling.”

It’s the perfect starting point because it sits at the intersection of several factors:

  • High volume — Millions of homeowners face this annually
  • Real urgency — In hot climates, this is a summer emergency
  • Diagnostic potential — There are 8–10 meaningful diagnostic paths to explore
  • Mixed DIY/Pro boundary — Some checks are safe and useful; others require a technician
  • Genuine confusion — Most homeowners don’t know where DIY ends and professional work begins

The long-term vision is broader. Over time, the platform could extend to plumbing, electrical triage, appliances, and general home maintenance. But I’m not there yet.

Start with one golden path. Solve it obsessively well. Then expand.

That discipline is what separates products that grow thoughtfully from products that sprawl.

Why Architecture Matters Here

This isn’t “build a chatbot and ship it.” And that’s intentional.

XpertConnect™️ is being built with a .NET-first, domain-driven architecture because the product needs structure, not just clever AI prompts.

The system has to support:

  • Durable session state — Users might troubleshoot over multiple days
  • Deterministic decision logic — The same inputs should always produce the same step
  • Safety boundaries — Built into the system, not dependent on AI behavior
  • AI integration — As a support layer, not the core decision-maker
  • Future marketplace workflows — When escalation leads to booking or coordination
  • Long-term maintainability — Code that teams can reason about and evolve

If I cut corners on architecture now, I’ll spend the next year fighting technical debt instead of building a better product.

The Layered Shape

The current architectural direction is a layered design:

Domain Layer — The core business concepts and rules. What makes a troubleshooting step safe? What conditions trigger escalation? These are domain decisions, not infrastructure decisions.

Application Layer — Use case orchestration. How do we move a user through the troubleshooting flow? What happens after they answer a diagnostic question? This layer speaks the language of the product.

Infrastructure Layer — Persistence, AI service calls, external integrations. Where does session state live? How do we call OpenAI? Where do we store user data?

Web/API Layer — The experience layer. How do we present this to users on the web? How do we expose future mobile or third-party integrations?

This separation gives us immediate benefits:

  • Clearer reasoning — Each layer has a single responsibility
  • Better testability — We can test domain logic without infrastructure
  • Less accidental coupling — Changes to the database don’t cascade everywhere
  • Confidence in expansion — We can add features without everything becoming spaghetti

The Orchestrator Layer

One of the core product ideas is that troubleshooting should be structured, not improvised.

That means the system doesn’t just ask AI, “What should the user do next?” Instead, it follows a predetermined progression:

  1. Evaluate the current state
  2. Determine what the next valid step is
  3. Check safety boundaries
  4. Pause if the user needs to acknowledge risk
  5. Escalate if necessary
  6. Otherwise, present the next diagnostic step

This orchestration layer is critical because the product logic owns the flow. AI supports the experience inside those boundaries.

That distinction changes everything. It means:

  • The system can guarantee certain safety properties
  • Users follow a consistent, predictable flow
  • We can A/B test individual steps
  • AI improves the experience without controlling the workflow

Most AI products get this backwards. They let the model decide what to do next. I’m doing the opposite.

What AI Actually Does Here

AI is important to XpertConnect™️. But only when it’s given the right job.

The AI layer has two core responsibilities:

1. Normalize messy user input — Users describe problems in unpredictable ways. “It’s like making a grinding noise but sometimes it doesn’t” or “It just stopped working, I guess.” The AI layer takes that unstructured description and normalizes it into structured context the system can use.

2. Generate contextual guidance — The system knows what step to present. The AI makes that step feel natural and specific to the user’s situation, rather than generic.

That’s fundamentally different from asking a model, “What should the user do next?”

The goal is clear: make AI useful without making it the decision-maker.

In practice:

  • The AI gets explicit context about what it’s being asked to help with
  • The AI operates within defined constraints (safety rules, flow boundaries)
  • The deterministic system still owns critical decisions
  • The user experience is better because of AI, not dependent on it

That’s how you build AI products people actually trust.

What I’m Not Sharing Publicly

This post intentionally excludes several things that belong in private documentation, not a blog:

  • Internal state transition rule design
  • Domain entity counts or implementation status checklists
  • Audit logging specifics (IP capture, device fingerprinting, etc.)
  • Detailed code snippets that reveal architectural internals
  • Exact persistence mechanics or read-model strategies
  • Claims better kept in engineering documentation

This isn’t about being secretive. It’s about being smart.

Public readers need the vision, philosophy, and architectural discipline. They don’t need the details that should stay private.

The Current Phase

Right now, the focus is on building the right foundation.

That includes:

  • Clarifying the domain model — What are the core concepts? What rules govern transitions?
  • Defining safety boundaries — Where does DIY end? What triggers escalation?
  • Shaping the orchestration layer — How do we move users through the flow?
  • Integrating AI meaningfully — Where does AI add value without controlling outcomes?
  • Planning for scale — How will this evolve beyond HVAC?

This phase isn’t flashy. It’s architectural work that determines whether the product feels solid and trustworthy or held together with tape.

Most projects skip this, rush to features, and regret it. I’m not doing that.

Why Build This At All

After two decades in software, you get good at spotting problems that are both technically interesting and genuinely worth solving.

This is one of them.

The information for safe DIY troubleshooting exists. It’s just buried, scattered across forums and support docs, often generic and contradictory. People either give up and call someone immediately (expensive), guess their way through (risky), or spend hours on YouTube (frustrating).

I want to build a platform that feels like this:

  • Calm — No overwhelming options or endless conversation
  • Structured — Clear progression, not rabbit holes
  • Safe — Explicit about what you should and shouldn’t try
  • Genuinely useful — Better than the alternative
  • Honest — Clear about where professionals take over

That’s the product underneath all the architecture.

Where This Could Go

HVAC is the beginning, not the end.

If the model works — if users feel safer, more confident, and actually solve their problems — the same pattern scales:

  • Plumbing diagnostics
  • Electrical triage (identifying when you need a licensed electrician)
  • Appliance troubleshooting
  • Seasonal home maintenance
  • Eventually, a marketplace connecting users with vetted service providers

But the rule is simple: nail the first thing before expanding.

One golden path. One high-value problem. One experience that’s demonstrably better than the alternatives.

The Real North Star

The best AI products don’t just answer questions faster.

They reduce uncertainty.

For XpertConnect™️, that’s the mission: help people move from confusion to clarity — safely, step by step, with a real human handoff when the situation calls for it.

That’s the build. And I’m excited to see where it goes as it evolves.

Want to follow along? Check the What’s New page for project updates, read the architecture deep dive for more technical details, or reach out if you want to discuss .NET architecture, AI product design, or building in public.