[ OFFER ARCHITECTURE ]

Triage. Rescue. Fractional FDE.

[ CAPABILITIES ]

What we build, and what we refuse to build.

Six capability surfaces. Each one ships with a named failure mode, an eval, and a runbook.

[ CASE FILES INDEX ]

Every case file, on one page.

[ FAQ ]

The questions buyers actually ask.

Pricing, scope, refusal criteria, eval methodology. Answered in writing so we do not repeat them on every call.

• How much do your engagements cost?+

  Three tiers, all fixed-fee. Production Triage is $5,000 for a 7-day architecture review and go/no-go call on a stuck Claude pilot. Workflow Rescue is $25,000 to $40,000 for a 21-day fixed-scope production deployment of one stuck workflow. Fractional FDE is $25,000 to $40,000 per month with a 3-month minimum, structured as an embedded Forward Deployed Engineer model. We do not bill hourly, we do not run open-ended retainers, and we do not negotiate scope in arrears. Each engagement has a written scope and a written outcome up front.

• What is a Production Triage and why does it cost $5K?+

  Seven days, $5,000, one written deliverable: an architecture review of your existing Claude pilot, a sequenced 21-day fix list, and a go/no-go call on whether the pilot is worth shipping. We read the code, the prompts, the logs, the MCP wiring, and the eval suite (or note its absence). The price is the qualification gate. It filters out projects that need a Workflow Rescue from projects that need a different vendor or a different decision, and it gives both sides written evidence to commit to or walk away from. 50% of AI pilots fail at the org layer, not the tech layer.

• Who is your ideal client?+

  An engineering leader, CTO, or AI task-force owner whose Claude pilot is stuck between demo-good and production-ready. Three weeks from a deadline, or six months past one. Existing in-house engineering capability that can own the system after handoff. US-based for billing and timezone alignment, primarily mid-market SaaS, enterprise IT, fintech, and operations-heavy verticals. We are not a fit for greenfield prototypes that have not been built yet (that is your team's job) or for clients looking for automation-consultancy strategy decks.

• What counts as a blocked pilot?+

  A Claude or agentic pilot that demos cleanly but cannot move to production. Common shapes: MCP integrations are leaking permissions or burning tokens; the eval suite does not exist or has not been run on real traffic; observability is absent so on-call cannot triage; the architecture cannot scale past the demo entity count; the team built it once and cannot extend it without the original contractor. We have a written triage checklist; if your pilot matches three or more lines on it, Production Triage is the right next step.

• When will you refuse a project?+

  When the pilot has no salvageable architecture. When the ICP is wrong (consumer, low-trust, or no in-house engineering team to own the handoff). When the timeline is shorter than the fixed scope and there is no path to extend it. When the budget cannot cover the named outcome and the conversation is about discounting rather than narrowing scope. We say no out loud and we say it in writing. Refusal is part of the offer. We rescue the half worth saving and refuse the rest.

• What is in scope for a Workflow Rescue?+

  Fixed-scope 21-day production deployment of one stuck workflow. In scope: prompts, retrieval, tools, guardrails, hardened MCP integrations with scoped auth and rate limits and audit logs and output validation, regression eval suite on real traffic, observability wired to your existing stack, runbook, pair-programming handoff to the owning engineer, architecture doc, infrastructure-as-code, every secret in your vault. Out of scope: a second workflow (that is a second Rescue), long-tail maintenance (that is Fractional FDE), org-design or change-management consulting.

• What does Anthropic Academy certified actually mean?+

  Nine Anthropic Academy completions to date: Agent Engineering, Claude Code, MCP, Context Engineering, and five additional certificates. CCA-F (Claude Certified Architect, Foundations) is pending. Each certificate is a coursework completion with a written or applied component validated by Anthropic. The coursework covers the same agent patterns and failure-mode taxonomy we ship inside Workflow Rescue and Fractional FDE engagements. Certifications are necessary, not sufficient. They get you in the door; the case files are what close the door behind you.

• Are you Columbus-only?+

  Headquartered in Columbus, Ohio. Engagements are US-based for billing and timezone alignment. Workflow Rescue and Fractional FDE engagements are run remote-first with optional onsite weeks negotiated case by case. Production Triage is fully remote. Local Columbus and central Ohio clients are welcome; we are not exclusive to the region. Our 13-year US enterprise IT background (Wendy's Help Desk, OhioHealth, Sedgwick, Blair IT) is part of why we are based here.

• How do you build production MCP servers?+

  Two production MCP servers shipped to date. Patterns: scoped OAuth, typed tools, rate limiting, audit logging, output validation on every response. We default to read-mostly scopes with explicit write-confirmation gates. Failure mode named: MCP transport latency under high concurrency is the dominant tail-latency contributor, mitigated by tool-level timeouts and a circuit breaker. Handoff package: deployment runbook, eval suite, written threat model. The same hardening practice ships inside every Workflow Rescue and informs the standards doc left behind by every Fractional FDE engagement.

• What is your eval methodology?+

  Regression evals on real traffic before launch, sampled cadence after launch. The suite covers the happy path, the named failure modes, and adversarial cases pulled from your incident history. We do not ship vibes; we ship the eval report. Failure mode named: eval drift when the underlying model version changes, mitigated by pinning the production model and gating upgrades through a re-run of the full suite. Up to 90% pre-launch confidence on the bounded scope. Out-of-scope inputs are routed to the exception queue, never auto-handled.

• Why Fractional FDE instead of a fixed Rescue?+

  Workflow Rescue gives you one workflow in production in 21 days. Fractional FDE gives you ongoing senior AI implementation muscle across three months: pair programming, design reviews, MCP and Claude Code training, two to three additional production workflows shipped alongside your engineers, written standards docs, hiring-loop support. You leave with senior in-house capability, not dependency. Best fit after one or two successful Rescues, or when the team needs to build agentic muscle internally rather than buy a single workflow. Each month has a written work plan to prevent scope creep.

• What providers and tools do you operate?+

  Multi-provider by default. 2.27 billion OpenAI tokens and 559 Codex sessions in production. 30+ Claude Code skills shipped. Two production MCP servers. Single implementations utilizing upwards of 50 concurrent worker nodes when the workload supports it. Stack defaults: Anthropic, Claude Code, MCP, Lovable, Supabase, n8n, Python, TypeScript, React, Tailwind, Vite. Vendor lock-in is a named failure mode we design against; orchestration layers are written to be portable across providers.

[ THE OPERATOR ]

Alex Cinovoj. Columbus, OH.

13 years US enterprise IT (Wendy's Help Desk, OhioHealth, Sedgwick, Blair IT, TechTide AI Oct 2024). Lovable Senior Champion with 10,183 verified edits across 277 active days. 9× Anthropic Academy completions, CCA-F pending. Co-founder Persyn.ai (with Anette Benson). Co-founder FigGlow.ai (with Shane Spencer). 56,424 LinkedIn followers. Top 5 AI Education, Favikon, November 2025.