Real engagements. Real outcomes.

Problem

Divorce cases require both a calculation (separation math across asset classes) and a valuation (what those assets are actually worth). Clients often leave initial consultations without a clear picture of what they’re entitled to. Generic LLMs are a bad fit. They hallucinate numbers and over-estimate. Determinism matters more than fluency.

Approach

Integrated the firm’s proprietary separation calculation into an app that gives clients a grounded estimate. Layered AI on the asset-valuation side, designed for conservative estimates. The goal is to avoid over-inflation that creates false expectations. Ongoing work expands scope as Carey identifies more places AI can augment (not replace) the firm’s legal judgment.

Outcomes

• →Active, expanding engagement
• →Moves consultations from "I don’t know what I’m entitled to" to "here’s a grounded estimate"
• →AI applied in a regulated domain without sacrificing accuracy or defensibility

Problem

Mid-size operations carry hidden friction in every department: repetitive data entry, broken handoffs between sales and operations, ad-hoc invoice processing, organizational knowledge buried in inboxes, historical data scattered across systems. The first job is finding where AI actually moves the needle versus where a simpler fix works better — and being honest about both.

Approach

Embedded across departments in learning mode first. Mapped workflows: where time goes, where mistakes happen, where judgment shows up. Built where AI fit: automations for repetitive work, internal AI chatbots for organizational knowledge, AI-augmented invoice processing. Upgraded database infrastructure to support multi-year historical queries. Each ship was the smallest version that proved the fit, then expanded from there.

Outcomes

• →AI tooling live across sales, operations, and construction departments
• →Automations handling repetitive work that previously required manual touch
• →Internal AI chatbots improving day-to-day work and research access
• →Database infrastructure upgraded for multi-year historical queries

Problem

The earlier build was broken in the places that mattered. Generic fitness apps collapse the nuance bodybuilders care about. "Incline DB press, 30°, neutral grip" is not the same exercise as a flat barbell bench. Serious lifters track them as distinct. On top of that, bodybuilding is social: lifters want to share, get feedback, meet others in the scene.

Approach

Started over from scratch in React Native. Designed a custom exercise matching system covering 180+ exercises across grip, stance, position, unilateral vs. bilateral, and cable vs. free weight, so every logged set is a distinct, analyzable variant. Built the social layer around real community behavior. Handed off with the patterns, context, and momentum for Jamie to take it across the finish line himself.

Outcomes

• →Stalled product restarted from scratch and shipped
• →Custom exercise matching across 180+ variants
• →Social layer designed around real bodybuilder behavior
• →Founder launched to the App Store himself, 5.0 ★ rating

Problem

The previous build was broken where it mattered most: MLS and IDX integration. For a niche realtor, filter logic is the product. "Homes near you" is meaningless when your clients are hunting for equestrian estates in a specific price tier.

Approach

Stabilized what was there. Rebuilt MLS/IDX integration with custom filter logic for equestrian properties, acreage, luxury tier. Designed the site around her brand so the first visit feels like the right realtor. Wired an automated lead workflow: visitor filters listings → contacts her → she gets notified → no leads dropped.

Outcomes

• →Site live at pastureandporch.com
• →Custom filter logic working against live MLS/IDX feeds
• →Automated contact-to-notification lead flow
• →Serving million-dollar home traffic in her niche

Problem

Engineers who know how to code don’t automatically know how to ship with agentic tools. The leverage is in environment setup, prompting patterns, guardrails, and knowing when to let an agent run versus when to take the wheel. Without that, the tools sit underused or produce code nobody wants to maintain.

Approach

Set up Claude Code and Codex with the right rules and project conventions. Walked through agent orchestration in real work — when to spawn multiple agents, when to constrain scope, how to review what an agent produced. Shipped one real app together as the proof point. Patterns transfer to any stack, any future project.

Outcomes

• →1-month engagement
• →Developer shipped his first agent-built app end to end
• →Transferable patterns for future projects on any stack
• →Foundation for ongoing relationship as new tools emerge