Capabilities.

We design and build production software with AI models embedded in the core workflow, not added as a chat widget: document understanding with page level citations, computer vision pipelines that compare jobsite photos to a plan, and voice guided task assistants. Every system we ship, we operate ourselves in production, so we carry direct knowledge of the operational cost and failure modes of the AI we build.

In production

Debriefed · Wheelingo · Cared AI

In pilot / demo

Dexter (vision) · Calyssa (clinical extraction)

Patterns

page cited Q&A · photo based VQA · structured extraction · voice guidance

Our production systems run on a mix of AWS, including EC2, Lambda, DynamoDB, API Gateway, Cognito, and S3, and Cloudflare, including Pages, Workers, and D1, chosen per workload rather than defaulting to one vendor. We build for cost visibility and contained blast radius from the start: per pool spend caps, signed origin authentication between edge and origin, and infrastructure a small team can operate without a dedicated operations function.

AWS surface

EC2 · Lambda · DynamoDB · API Gateway · Cognito · S3

Cloudflare surface

Pages · Workers · D1 · DNS/CDN

Doctrine

Per workload platform choice. Spend caps and blast radius limits from the first commit.

We move from a stated problem to a working, testable prototype quickly because every person on the team can design, build, and deploy independently, without waiting on a separate design or infrastructure group. The AI guided staff training platform on this site is one example: a working prototype built as internal research and development to validate an interaction pattern before proposing it to a customer.

Evidence

AI guided training platform prototype ↗

Method

One engineer owns problem, design, build, and deploy. No handoff chain.

Every production system we operate has been through internal adversarial security review covering authentication, per user and per IP rate limiting, resistance to prompt injection in AI facing inputs, and data isolation between accounts, with findings fixed before the next release rather than tracked and deferred. We treat all inputs, including inputs passed to AI models, as untrusted until verified.

Review scope

authn · rate limiting (per user, per IP) · prompt injection · data isolation

Posture

All inputs untrusted until verified, including inputs to AI models.

Discipline

Findings fixed before release, not deferred to a backlog.

As a small team that has personally written, deployed, and debugged the systems described on this site, we advise on architecture choices, AI model and vendor selection, and build versus buy tradeoffs from direct operating experience rather than from a slide deck. That includes plain assessment of what an AI system can and cannot reliably do in production.

Scope

Architecture review, AI model and vendor selection, build versus buy analysis.

Stance

Plain assessment of what AI can and cannot reliably do in production.

Future Vanguard LLC carries commercial general liability, professional software development errors and omissions, cyber liability, workers compensation and employers liability, and hired and non-owned automobile liability coverage. Forms, limits, carriers, and policy periods are listed below. These are insurance policies. They are not a surety bond, a certification, or any form of government approval.

General liability

Commercial general liability, occurrence form. Each occurrence $1,000,000. General aggregate $2,000,000. Products and completed operations aggregate $2,000,000. Damage to rented premises $50,000. Medical expense $5,000. Personal and advertising injury included.

Professional Software Development E&O

Errors and omissions coverage for professional software development services, claims-made form. $1,000,000 per occurrence, $1,000,000 aggregate.

Cyber liability

$250,000, written under the Professional Software Development E&O policy.

Workers comp

Workers compensation per statute. Employers liability: each accident $100,000, disease each employee $100,000, disease policy limit $500,000. Proprietor and member not excluded.

Hired and non-owned auto

Automobile liability for hired autos and non-owned autos only. Combined single limit $2,000,000 each accident. Bodily injury $100,000 per person and $300,000 per accident. Property damage $100,000 per accident. The company owns no vehicles.

Policy period

07/28/2026 to 07/28/2027 general liability, E&O, cyber, workers comp
08/04/2026 to 08/04/2027 hired and non-owned auto

Carriers

Berkshire Hathaway Direct Insurance Company for general liability. National Liability and Fire Insurance Company for Professional Software Development E&O, cyber, and workers compensation. Technology Risk Retention Group, Inc. for hired and non-owned auto.

Producers

biBERK · Corgi Insurance Services, Inc.

Certificates of insurance available on request.