{"who":{"name":"Kevin Fincel","role":"Forward Deployed AI Engineer","credential":"Founder, Geol.ai (https://geol.ai)","base":"Jacksonville Beach, FL · On-site to start, remote delivery","workingModel":"Two or three days on site at the start of an engagement, then build and deployment from Jacksonville Beach, FL. Training visits are planned per engagement. Clients mainly in the United States and Canada."},"thesis":"Kevin Fincel describes his work: I work with your team to understand where the work gets stuck, then build AI into the systems you already use. That might mean helping staff work through quote requests, checking a product catalog, or connecting information that lives in different tools. I handle the build and deployment, and help your team put it to use.","offers":[{"name":"AI Deployment Sprint","scope":["Fixed scope","2–3 weeks"],"summary":"Kevin Fincel: The sprint gives you a clear decision about what to build and why. I spend the first two or three days at your business, watching the work and talking with your team. Over the rest of the two-to-three-week engagement, I check the data and access requirements, identify the risks, and put together the build plan and ROI estimate. We agree on how to measure success before the build starts. The plan is yours to use, whether I build it or someone else does.","entryPoint":true},{"name":"Proof of Value → Production","scope":["Build and test","Ready for daily use","Scoped per engagement"],"summary":"Kevin Fincel: The prototype and deployment are a separately scoped build stage. I build the workflow we chose in the sprint and test it against your data. Getting it into production also means sorting out access, logging and the steps that need human approval. Most of this happens from my office, with a visit for training if that helps your team. We keep working until your team can run it.","entryPoint":false},{"name":"Fractional AI Lead","scope":["Monthly","Part-time technical leadership"],"summary":"Kevin Fincel: For teams that need an AI lead but aren't ready to hire one full time. I help make architecture and vendor decisions, review the work, and get the team comfortable running it. I can also build when you need another pair of hands.","entryPoint":false}],"evidence":[{"title":"Connecting AI to the tools an agency uses every day","headline":"84 TOOLS · 5 SELF-HOSTED SERVERS","status":"LIVE IN PRODUCTION","url":"https://finceldesign.com/casework/five-mcp-servers-one-token-vault"},{"title":"Replacing a 12-tab spreadsheet with a shared operations app","headline":"92 MCP TOOLS · HAND-BUILT OAUTH 2.1","status":"LIVE IN PRODUCTION · DAILY TEAM USE","url":"https://finceldesign.com/casework/one-database-two-front-doors"},{"title":"Auditing 2,014 domains for tracking and consent problems","headline":"2,014 DOMAINS PROBED · 27 LIVE · TRIAGED","status":"AUDIT DELIVERED","url":"https://finceldesign.com/casework/2014-domains-one-consent-gate"},{"title":"Connecting a 59,886-item inventory catalog to a new storefront","headline":"5.4 MIN FULL SYNC · 59,886 ITEMS · 100% HASH-STABLE","status":"BACKEND VERIFIED · UI IN PROGRESS","url":"https://finceldesign.com/casework/live-erp-api-first-storefront"},{"title":"A 916-product catalog review the catalog owner could run","headline":"916 PRODUCTS · OPERATOR COMPLETED THE REVIEW","status":"SYSTEM DELIVERED · COMPREHENSIVE RUN COMPLETED BY THE OPERATOR","url":"https://finceldesign.com/casework/916-products-30-agents-one-operator"},{"title":"AI-assisted ad changes with approval before every change","headline":"35 LIVE MUTATIONS · EVERY ONE APPROVED & LOGGED","status":"IN PRODUCTION · ~18 ACCOUNTS","url":"https://finceldesign.com/casework/agent-written-ad-changes-safely"},{"title":"A 697-product pilot for under $2 in compute","headline":"<$2 COMPUTE · 697 PRODUCTS · 98.1% VERIFIER PAIR AGREEMENT","status":"PILOT EXECUTED & VERIFIED · PROGRAM PHASES GATED","url":"https://finceldesign.com/casework/23658-products-two-dollars-compute"}],"policy":"Client and employer names are withheld. Kevin can walk you through the work and discuss references in a conversation. Figures were checked against source repositories and project records in August 2026.","fitAssessment":{"url":"https://finceldesign.com/fit","method":"Local analysis distinguishes demonstrated delivery, transferable skills with a concrete ramp-up, and requirements needing discussion.","evaluationRecord":"https://finceldesign.com/evals"},"nextStep":{"capabilities":{"deliveryMode":"legacy","configured":true,"retrySupported":false,"enabled":true,"rate":"3 attempts per hour in the strict bucket. Agent requests need no spam check. Send one introduction per person you represent. Limits count HTTP attempts, including invalid requests, per server instance and trusted IP attribution bucket. IPv4 uses one bucket; IPv6 shares a /64 bucket. Unknown attribution uses one shared strict bucket. Shared networks share the allowance. These are request limits, not successful delivery counts.","retry":"Legacy mode: retrySupported:false. Idempotency-Key is ignored; no UUID validation or deduplication is active. A 504 may already have reached the provider; do not resend automatically."},"action":"POST","endpoint":"https://finceldesign.com/api/contact","schema":"GET the endpoint for the field schema","openapi":"https://finceldesign.com/openapi.json","note":"GET /api/contact for current availability and limits; /openapi.json documents submitIntroduction and machine-code retry decisions. Legacy mode: retrySupported:false. Idempotency-Key is ignored; no UUID validation or deduplication is active. A 504 may already have reached the provider; do not resend automatically. A successful response means provider acceptance, not inbox delivery or reading. No email address or phone number is published.","messageHint":"Describe the work that is getting stuck and the systems your team uses. Kevin will tell you whether AI could help.","rateLimit":"3 attempts per hour in the strict bucket. Agent requests need no spam check. Send one introduction per person you represent. Limits count HTTP attempts, including invalid requests, per server instance and trusted IP attribution bucket. IPv4 uses one bucket; IPv6 shares a /64 bucket. Unknown attribution uses one shared strict bucket. Shared networks share the allowance. These are request limits, not successful delivery counts. GET /api/contact for browser allowances."}}