Selected work

Real systems, real numbers, honestly labeled.

Client work is marked as client work. Concept builds are marked as concept builds, with modeled numbers. We'd rather show a smaller true number than a bigger invented one.

Economic Dev.AI Automation

Supply SA cut vendor certification from 30–45 min of staff time to a self-serve flow.

A production AI assistant with deterministic branching logic compresses a 30–40 step certification process into 7–11 screens. Vendors complete certification without staff intervention; reviewers step in only on edge cases.

~34
hrs staff time freed / month
$11.5K+
annual capacity recovered
Read the case study →
Property Mgmt.Ops DashboardConcept Build

A concept build: replacing six spreadsheets with one live rental portfolio dashboard.

A relational Airtable base paired with a live Bolt dashboard, modeled to reclaim hours a week otherwise lost to manual tracking.

~6 hrs
reclaimed / wk (modeled)
~310 hrs
redirected / yr (modeled)
Read the case study →
Knowledge BaseAI AssistantConcept Build

A concept build: turning a company's own documents into an assistant that answers in seconds.

A retrieval assistant over a company's document library. Files are chunked, embedded, and searched with a reranking step, so answers stay grounded in the source material instead of guessed.

Seconds
to a sourced answer vs. manual searching
1 drop
a new file updates the whole knowledge base
Read the case study →
Lead GenAI AutomationConcept Build

A concept build: qualifying, scoring, and nurturing every inbound lead before a human opens the record.

A no-code n8n system that qualifies each lead, saves the rest instead of discarding them, and hands every good one a personalized report plus a week of follow-up. Modeled to redirect ~25 hours a month from manual triage.

~25 hrs
redirected / mo (modeled)
$625/mo
modeled capacity recovered
Read the case study →
Outbound SalesAI AutomationConcept Build

A concept build: recovering the half of a lead scrape that arrives with no email at all.

Two connected n8n workflows turn a raw Apollo/Apify scrape into personalized cold email drafts. A two-tier lookup recovers the leads the scrape missed, and every attempt is free unless it lands.

2 tiers
LinkedIn URL, then company domain
$0
cost per failed lookup
Read the case study →
Social MediaContent EngineConcept Build

A concept build: turning an idle lead list into three platforms of content grounded in each lead's own site.

A two-part n8n system that scrapes every lead's site, turns what those companies actually say into five angles, and writes LinkedIn, Facebook, and Instagram versions in parallel once a human approves one.

~20 hrs
research & drafting saved / batch (modeled)
3 platforms
written in parallel from one approved angle
Read the case study →
Outbound SalesAI AutomationConcept Build

Tested on 270 real leads: cold emails grounded in the lead's own site, with zero invented details.

Every lead's site is scraped before the model writes a word, so drafts reference something real instead of plausible fiction. A scrape that failed used to feed the model its own error message. That bug got fixed at the root, not papered over with a better prompt.

500→270
raw leads cleaned to verified-email leads
Zero
invented details across the tested batch
Read the case study →
Sales OpsAI AutomationConcept Build

A concept build: a finished discovery call becomes a branded PDF proposal with nobody re-listening to the recording.

Claude drafts four proposal sections from the call transcript itself, so it can't quote a number that never came up. A code node builds the HTML instead of a second AI call, which removed an entire category of failure at no added cost.

14/14
nodes passed on an end-to-end test run
~6–8 hrs
redirected / mo (modeled)
Read the case study →

Your turn

The next case study could be your Tuesday.

Soulwire Studio

Systems with soul: custom websites, AI automation, and the digital ecosystems behind them, built for mission-driven businesses.

© 2026 Soulwire Studio · Veteran-owned · San Antonio, TX