SOPHIA XT · Investor Pitch · 2026
Operator-built AI implementation,
not the consultancy version of it.
We run service businesses, ship the operational software they need, and publish the research underneath it. Each of those three is checkable: the trucks, the deployed systems at /work, and the Apache 2.0 repositories at /releases.
systems in daily use
repositories, Apache 2.0
papers and reports
held-out advantage
Each of these opens to the thing that produced it: the systems are listed with a URL each at work, the repositories at releases, the papers at papers and downloads, and the masking result with its 95% interval [1.51, 1.85] at anchor-token masking.
Operator, builder and researcher, all three provable.
Most AI implementation companies ship demos, not durable workflow change. SophiaXT is different in three load-bearing ways.
We operate
We run real service businesses (The Appliance Guys, HVAC Guys, Becker Service Group). We know the workflow before we touch a model.
We build
We ship the operational software we'd want — Moe intake, DiagBuddy diagnostics, Sophia Key 3M cold-chain decision support, Lead Leak Scanner.
We research
We publish the work with its experiment log. Cassandra T1 is an open masked-diffusion prototype on GitHub and the anchor-token study behind it reports a 1.67x held-out advantage at [1.51, 1.85].
Operator, builder and researcher under one roof. Whether that is a moat is for an investor to judge; that all three are running is something they can check in an afternoon.
Service operators are getting pitched fantasies.
Workflow ≠ chatbot
Most AI projects fail because they automate the wrong step.
Vendor lock-in is expensive
Off-the-shelf SaaS doesn't bend to how a real service company actually books, dispatches, diagnoses, and invoices.
Owners can't tell who's legit
A 3-month "AI strategy engagement" produces slides — not a working system.
Most consultants haven't run a truck
They've never lost a $400 job because dispatch was 22 minutes late.
A workflow-first AI implementation company.
Audit before recommending
We inspect the actual workflow first — and tell the client when AI is unnecessary.
Build the smallest useful system
Automation, API wrappers, dashboards, custom models. Only where the math works.
Own the operating context
We run the same kinds of businesses we serve. Recommendations have to survive on our own dispatch board first.
Publish the research
Customers and partners can verify our depth. Cassandra T1 is public on GitHub today.
Live product surface, not a roadmap.
Every entry below is verifiable in a screen share. URLs are real and indexed.
SophiaXT.com
Implementation services site, Models page, Lead Leak Scanner, Hostinger Node deployment.
ServiceLens
Appliance diagnostic SaaS — paid users, Stripe billing, Encompass + Marcone parts, OpenAI + Q3M dual-provider.
Fix Appliances Now
The Appliance Guys repair business — Moe AI intake chatbot, GE dealer catalog, admin panel.
Cassandra T1
Open masked-diffusion edge model — architecture, scheduler, evaluation runner public on GitHub.
Sophia Key 3M
Cold-chain shipment simulator — thermal solver, route + weather modeling, decision brief generator.
Lead Leak Scanner
Free MVP audit tool — server-side Q3M analysis, themed PDF report, top-of-funnel for inbound calls.
Moe (Intake)
Sophia/Mercury-powered chatbot for appliance repair intake. Live at fixappliancesnow.com.
AI Ad Creator
Sophia copy + fal.ai image-generation pipeline for ad creative across owned brands.
The numbers are real. They are also small.
We are early. We are showing the actual signal so an investor can verify it themselves rather than believe a deck.
| Asset | Status | Verifiable signal |
|---|---|---|
| DiagBuddy | Live | Paying users on Stripe billing, company accounts, Encompass and Marcone parts. Exact counts come off the dashboard on a call; none is published here. |
| The Appliance Guys | Live | Service trucks running · HouseCallPro bookings · Synchrony + Wisetack live |
| sophiaxt.com | Live | Indexed, schema deployed, 24 papers and 9 public repositories listed |
| Cassandra T1 | Public release | GitHub repo public · architecture docs · evaluation runner · epoch-5 checkpoint with its SHA-256 |
| Anchor-token masking | Published | 1.67x held-out advantage, 95% interval [1.51, 1.85], paper and technical report both downloadable |
| The HVAC Guys | In build | Same playbook as Appliance Guys, separate domain |
| Sophia Key 3M | Working prototype | Thermal solver + route + decision brief functional |
No user count, ARR figure or letter of intent appears on this page. A user count that cannot be checked from outside is worth less than the dashboard it came from, so a discovery call gets read-only access to the dashboard instead.
Thomas Garren
Founder & Chief AI Architect
Thomas runs the company alongside operating responsibilities at Becker Service Group (with Chris Becker). The operating background — appliance repair, HVAC, mobile field service — is exactly the customer profile SophiaXT serves. Recommendations have to survive on our own dispatch board before they're shipped to anyone else's.
He is a live service-business operator, a shipping product owner and a publishing researcher at the same time, and that combination is the reason SophiaXT exists at all.
Full biographyThe Appliance Guys
Live service business running real trucks. HouseCallPro booking, Synchrony + Wisetack financing, GE dealer status.
The HVAC Guys
In build. Same operating playbook as Appliance Guys, separate domain, shared service group.
Becker Service Group
Parent operating layer with Chris Becker. Provides the field-service context behind every SophiaXT engagement.
Beachhead first. Then expand.
We are not chasing the entire AI market. The beachhead is technical service businesses in the United States — appliance repair, HVAC, plumbing, electrical, mobile field service, technical logistics, cold-chain.
These are owner-operator businesses with real revenue, weak software, and a deep aversion to hype. They are also the businesses we already operate, which makes the sales motion natural rather than forced.
Channel 01
Self-operated
Our own service businesses are reference customers, video case studies, and a constant source of feature ideas. We are customer #1.
Channel 02
Implementation
Paid engagements with other operators who recognize themselves in our story. Lead Leak Scanner is the top of this funnel today.
Three revenue lines, in the order we believe they scale.
now
Implementation Services
Workflow audit, fixed-fee build engagements, ongoing development retainers. Where most current revenue lives.
now → growing
Productized Software
DiagBuddy SaaS subscriptions ($20–34.99/mo, grandfathered pricing for early users). Lead-magnet tooling. Owned operating businesses generating cash.
medium term
Custom Model & Research
Custom model design, fine-tuning, and Cassandra-line research deliverables for partners interested in edge / diffusion-based models.
The cross-sell is real: a customer who hires us for an implementation often becomes a DiagBuddy account, a model-design engagement, and a long-term retainer.
What we are not.
| Generic AI consultancy | SophiaXT |
|---|---|
| Slides, no operating context | We run service trucks ourselves |
| Every workflow needs AI | We tell clients when AI is unnecessary |
| Long discovery → long build | Smallest useful system first |
| Closed methodology | Public research (Cassandra T1) |
| Vendor-locked SaaS | Multi-provider, OpenAI / Anthropic / Q3M |
| Quarterly retainer with no exit | Fixed-fee pilots with measurable outcome |
Research-only AI startups are not our peer either: most do not ship operating software for paying small businesses. We do, with revenue.
LTMi-XT — open infrastructure that compounds every customer engagement.
Most AI consultancies build one-off systems for each client and the IP dies in a private repo. SOPHIA XT's customer projects feed into LTMi-XT, an open knowledge-architecture stack we already ship under Apache 2.0. Every engagement leaves us — and the broader market — with reusable infrastructure.
Reference architecture
Closed knowledge layer
- · Closed atomization format
- · Hosted-only beta with sign-up gate
- · Retrieval-only consumer
- · No published reference implementation
- · No verifiable runnable demo from the open web
LTMi-XT (us)
Open + dual-purpose
- · Open
.ltmifile-format spec, Apache 2.0 - · Live demo at
sophiaxt.com/tools/ltmi-xt— no signup - · Same artifact serves retrieval and fine-tune ingestion
- · Reference TypeScript impl, CLI, web UI, Dockerfile on GitHub
- · Verified working end-to-end on production right now
For investors, this means
Customer dollars produce two assets at once: the deliverable for the client AND a reinforcing layer of the LTMi-XT open stack. Every implementation extends our public surface area.
For customers, this means
No vendor lock-in. The format is open and inspectable. If SOPHIA XT disappeared, the customer's .ltmi bundles still parse and still feed any retrieval or fine-tune pipeline they want to use.
For LLM platforms, this means
A reusable structured-knowledge primitive that solves both the retrieval-grounding and the fine-tune-data-prep problems. SOPHIA XT positions naturally as the company behind the format.
Verifiable in 60 seconds
We don't ask anyone to take this on faith. Live API at sophiaxt.com/api/ltmi-xt/crystallize accepts a POST with messy text and returns a structured .ltmi bundle with breadcrumb-anchored loci, lattice coordinates, and provenance. Reference implementation at github.com/Chorozion/LTMi-XT runs end-to-end with one npm install. We tested it before writing this slide.
The combination matters: an operating service-business shop that publishes its own AI infrastructure as an open, dual-purpose format. The format, its reference implementation and its licence are public, so a reader can decide for themselves how hard that is to copy.
The things that could go wrong.
Founder concentration
Thomas is core to product, research, and operating context. Mitigation: hires this round, plus the SophiaXT site itself functions as a living methodology document.
Service-business sales cycles are slow
Trust is earned in the field. Mitigation: free Lead Leak Scanner / audit funnel, plus our own operating businesses as reference accounts.
AI vendor pricing volatility
OpenAI / Anthropic / Inception cost changes affect implementation margins. Mitigation: multi-provider architecture, plus owning more of the stack via Cassandra-line research.
Cassandra T1 is a prototype
It is intentionally positioned as a public proof-of-concept. We have not invented benchmark numbers and we do not claim production readiness.
Multi-product surface area
Two operating businesses, a SaaS, a consultancy and a research line is a lot. Mitigation: investor money buys real headcount per line, not new projects.
What this round buys.
A single objective: 18–24 months of credible execution that proves the operator-built AI implementation thesis at one or two named customer accounts plus continued DiagBuddy growth.
Engineering hires
Two senior full-stack/AI engineers to absorb the implementation backlog and ship DiagBuddy v2 cleanly.
Sales / partnership lead
An operator-friendly seller who can run discovery calls in service-business vernacular.
Production hardening
Full observability, multi-region resilience, automated deployment pipelines, structured incident response.
Cassandra-line research
Modest compute budget for the next training milestone and a published technical paper.
Operating runway
18–24 months across SophiaXT and DiagBuddy with explicit per-line headcount.
We will share the specific dollar amount and per-line use-of-funds breakdown on a live call, alongside current bank statements and the DiagBuddy admin dashboard.
The timing is honest.
The AI consultancy market is overcrowded with non-operators, and service-business owners have already been burned. They are now actively shopping for someone who actually gets it.
LLM and diffusion-model costs have dropped to where a tight workflow plus a small custom model can outperform a much larger SaaS. That favors implementation specialists over platform incumbents.
Field-service workforce demographics (boomer retirement) are forcing diagnostic + dispatch automation. It is no longer optional.
Open research (Cassandra-line) compounds reputation in service-operator and technical communities, not just AI Twitter.
A 30-minute call gets you everything below.
Let's talk.
Email Thomas directly. Or use the contact page to book a discovery call on the calendar.
SOPHIA XT · St. Louis, MO · 2026