Cookies · Your choice

Analytics cookies, only if you say so

This site uses Google Analytics to count visits and see which pages get read. Those cookies stay off until you allow them. Everything the site needs to work runs without them. Cookie Policy.

This site uses Google Analytics to count visits. You can opt out at any time on Your Privacy Choices. Cookie Policy · Your Privacy Choices

Field dossierRev. 2026

Kevin FincelI put AI to work in your business.

Forward Deployed AI Engineer · Founder, Geol.ai · Jacksonville Beach → On-site → Remote

I help you figure out where AI would be useful, then build it into the systems your team already uses. That might mean checking a product catalog or helping staff work through incoming quote requests. I start with two or three days at your business, do the technical work from my office, and stay involved until people are using it.

  • Business automation
  • Connected systems
  • AI tools for your team
Hiring for an FDE role? Start here.
Kevin Fincel at his desk, about to speak
The film · 2 min 46 sFILM-01
§ 01

What makes this work

Strategy is cheap.
Shipping is not.

The hard part often starts after the first demo. Your records disagree, the software can't access what it needs, or the person doing the work doesn't trust the answer. That's where I spend my time.

Before I build, I learn how the job gets done now. We decide what the software can change on its own and what needs your team's approval. Then we test it. AI will get things wrong, and catching those mistakes is part of the work.

I've been responsible for getting business systems into daily use for more than eighteen years. The tools have changed. I'm still accountable for whether the thing works.

Shipped web & AI systems
18+ yrs
Shipped web & AI systems
Sales → delivery → P&L
3× founder
Sales → delivery → P&L
From the first call to daily use
Hands-on
From the first call to daily use
§ 02

Working with me

We start with one job your team needs to do better. The first engagement is a short, focused project to decide whether AI is worth building for it. You don't need to know which tools to choose. That's my part.

Phase 1 · Start here

AI Deployment Sprint

  • Fixed scope
  • 2–3 weeks
  • From $12K

The sprint starts with two or three days at your business, watching how the work gets done and talking with the people who do it. I take that back to my office and spend the rest of the 2–3 weeks checking the data the workflow depends on and working out the plan. We agree on how to measure success before committing to a build. You leave with a system map, the risks, a build plan and an ROI estimate. You can take that plan to someone else.

Phase 2

Proof of Value → Production

  • Build and test
  • Ready for daily use
  • Scoped per engagement

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.

Ongoing

Fractional AI Lead

  • Monthly
  • Part-time technical leadership

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.

I won't sell you an AI build if the problem needs a simpler fix.

§ 03

Selected work

These are projects I've built. I withhold client and employer names; we can discuss the work and check references in a conversation. The figures were checked against the source repositories in August 2026.

Case 01

A B2B marketing agency serving industrial manufacturers

Five MCP servers, one encrypted token vault

I connected the team's AI tools to the software they use for ads, reporting and calls. Staff sign in with their own accounts, and the ad connection can't change live spending.

84 TOOLS · 5 SELF-HOSTED SERVERS

  • MCP / FastMCP
  • Cloud Run
  • OAuth 2.1 + DCR
  • Firestore + Fernet
  • Secret Manager
  • Python
Shipped

Case 02

The same agency; internal operations platform

One database, two front doors

I replaced a 12-tab trade-show spreadsheet with an app the team uses daily. Staff can update it directly or ask an AI assistant to do it. Every change is recorded.

92 MCP TOOLS · HAND-BUILT OAUTH 2.1

  • Next.js 15
  • React 19
  • Supabase Postgres
  • MCP
  • OAuth 2.1
  • RLS
  • Vercel
  • Audit logging
Shipped

Case 03

A multi-brand industrial services group

2,014 domains, one consent gate

After years of acquisitions, the company didn't know which websites were still running. I checked 2,014 domains and identified where visitor tracking needed consent controls.

2,014 DOMAINS PROBED · 27 LIVE · TRIAGED

  • HTTP/DNS probing
  • Headless-browser verification
  • GDPR / CPRA
  • Consent gating
  • Event-driven JS
  • Severity triage
In field

Case 04

A multi-brand industrial electrical-equipment group

A live ERP, an API-first storefront

I built the connection between the company's inventory system and its new website. Nearly 60,000 items sync in 5.4 minutes. The backend is verified; the website is still being built.

5.4 MIN FULL SYNC · 59,886 ITEMS · 100% HASH-STABLE

  • Next.js
  • TypeScript
  • Hono API
  • PostgreSQL + Drizzle
  • NetSuite (OAuth2 M2M, ES256)
  • Azure Container Apps
  • GitHub OIDC
  • OpenAPI 3.1
  • MCP
In field

Case 05

An industrial safety-equipment distributor

916 products, 30 agents, one operator

I built an AI-assisted check of 916 products against a 250-page guide. The employee responsible for the catalog ran the full review herself, with three checkpoints together.

916 PRODUCTS · ~30-AGENT ADVERSARIAL SWARM

  • Multi-agent orchestration
  • Adversarial verification
  • PDF extraction
  • WooCommerce
  • Data QA
  • Human-in-the-loop
In field

Client and employer names withheld by policy. Every engagement verifiable in conversation.

§ 04

From the first call to daily use

The details change with the company. Here's how I usually work, and what you keep at each stage.

  1. 01

    Understand the work

    I spend two or three days at your business, talking to the people doing the work and tracing the systems behind it.

    You keep → A system map showing where the data lives and who can access it.

  2. 02

    Try it on your data

    We try one workflow on your data before committing to a larger build.

    You keep → A working prototype and a set of tests for its answers.

  3. 03

    Put it into use

    I set up access, approvals and monitoring so the team can use it safely and you can see what it costs.

    You keep → Instructions for running it and handling problems.

  4. 04

    Help the team use it

    I watch how people use it, fix what gets in their way and train the team, on site if that helps.

    You keep → Usage and outcome measurements we can compare with where we started.

  • D-01 · Build on what you already run. Replacing a working system needs a good reason.
  • D-02 · People approve the changes that are hard to undo. We agree on those limits before an agent gets access.
  • D-03 · Trust is earned in production. Test the answers and keep a record of what happened.
§ 05

Ask Kevin

AI answers based on my published work · You can check the sources

Ask KevinAI assistant

Ask Kevin · AI

Ask about the work Kevin has done, how an engagement starts, or what it would cost. Answers come from his published case studies, and every source is one click away.

Pressing Ask carries your question to the Ask Kevin page in this tab. Nothing is sent to a model until you ask there.

Standard · May use a hosted model · What this mode does

Your question and matching passage IDs go to this site's server, then to a hosted AI model. OpenRouter requests are routed only to providers that do not collect the request for their own use, so it is not used to train a model. Prompt logging is off on my OpenRouter account, so OpenRouter keeps the routing record and not your words. How long the provider that answers keeps it is set by their policy. Inspect run shows which model and provider answered. Please leave out confidential material.

The first question downloads about 58 MB of model and runtime files from this site. Your browser caches them for later questions.

Recorded example · 2026-09-10One question · 3 sources, word for word · No hosted model

Specimen · Recorded run · 2026-09-10 · Corpus b588a6f3 · No generative model · Local embedding + verbatim extraction

“We run Salesforce, three legacy databases, and our salespeople qualify RFQs by hand. What could Kevin build for us?”

EvidencedTop sim 0.663 · Floor 0.55

No generative model · Local embedding + verbatim extraction

  1. FAQ · What does Kevin Fincel deploy?Top sim 0.663

    What does Kevin Fincel deploy? Kevin builds AI agents and MCP servers, retrieval systems, and integrations with ERP, CRM, analytics and legacy software. The work includes testing answers, access controls and human approval for sensitive changes.

    Open the source →
  2. Case 06 · Agent-written changes to live ad accounts, safely · ProblemTop sim 0.627

    The team manages about 18 industrial-sector ad accounts. Agents can help with audits and campaign builds, but API changes affect live client spend. I needed a review process that made each change traceable and required a person to approve it before execution.

    Open the source →
  3. Case 04 · A live ERP, an API-first storefront · ApproachTop sim 0.618

    I separated the API, sync worker and Next.js storefront. The worker mirrors NetSuite into Postgres; the storefront reads through the API and never calls the ERP during a page request. A native part-number field was filled on 99.4% of sellable items. That gave us the grouping key for roughly 28,100 products with condition variants, without adding a client field. The sync fetches before writing, compares content hashes and refuses to reconcile a suspiciously small response. The database allows one runner at a time. CI checks OpenAPI drift, privileged routes fail closed, and a nine-tool read-only MCP server uses the same service layer as REST.

    Open the source →

This example was recorded at build time using the same model as the browser. It uses source passages word for word; no hosted model wrote the answer.

Open the full session →

Book the sprint with this conversation →

§ 06

Hiring for an FDE role?

If you're hiring a Forward Deployed Engineer, compare your role with the work I've done. The reader below links requirements to my case studies and flags where the experience doesn't match.

Fit readRuns in your browser

Will I fit your role?

Paste a job description to compare it with my work. The text stays in your browser.

0 / 40,000 chars · Your job description never leaves this tab
Or try a sample
resume.jsonresume.txtcontact
Agent-ready

If you're using an AI agent to research me, these endpoints give it my background, offers and case studies in a format it can read. It can also send an introduction through the contact API. The site is the demo.

Opens the file here so you can see what an agent reads.

§ 08

Field notes

Notes on systems I've built and decisions I had to make along the way.

2026

Claude Code → Claude Design: a research-first build playbook

How I use source code to check the facts before handing a project to a design tool.

2026

Five MCP servers, one OAuth vault

Architecture notes from self-hosting an agency's agent infrastructure on Cloud Run.

§ 09

About

Portrait of Kevin Fincel
K. Fincel · Jacksonville Beach, FL

I've been building for businesses since 2008.

I'm Kevin Fincel, an AI deployment engineer and three-time founder in Jacksonville Beach, Florida. For more than eighteen years, I've worked with businesses to understand how they operate and build the software they need. I'm used to being the person responsible when something breaks.

I lead AI infrastructure at a B2B marketing agency serving industrial manufacturers. I built its self-hosted MCP servers and the agent tools the team uses daily. I also founded Geol.ai, which measures how AI engines cite and recommend brands. Before that, my work covered full-stack web development, enterprise cloud and ERP integrations, often working directly with company leadership.

I've run workloads on Google Cloud (GCP) for five years and used Google Tag Manager (GTM) for six, including server-side tagging and consent-aware deployments. I've had Node.js (Node) services in production for two years.

Fincel Design, LLC is the company I founded in 2008. This site is what it grew into.

Published profile facts

  • Web and systems delivery: at least 18 years as of 2026-09 (attested 2026-09-02). Covers: web and systems delivery, web systems, full-stack development, full-stack product engineering.
  • TypeScript: at least 2 years as of 2026-09 (attested 2026-09-03). Covers: typescript.
  • Node.js: at least 2 years as of 2026-09 (attested 2026-09-04). Covers: node.js, node.
  • SQL: at least 8 years as of 2026-09 (attested 2026-09-03). Covers: sql.
  • Python: at least 6 years as of 2026-09 (attested 2026-09-03). Covers: python.
  • PHP: at least 9 years as of 2026-09 (attested 2026-09-03). Covers: php.
  • WordPress: at least 10 years as of 2026-09 (attested 2026-09-03). Covers: wordpress.
  • Docker: at least 2 years as of 2026-09 (attested 2026-09-03). Covers: docker.
  • AWS: at least 5 years as of 2026-09 (attested 2026-09-03). Covers: aws.
  • Google Cloud: at least 5 years as of 2026-09 (attested 2026-09-04). Covers: google cloud, gcp.
  • HubSpot: at least 3 years as of 2026-09 (attested 2026-09-03). Covers: hubspot.
  • Pardot: at least 4 years as of 2026-09 (attested 2026-09-03). Covers: pardot.
  • Google Tag Manager: at least 6 years as of 2026-09 (attested 2026-09-04). Covers: google tag manager, gtm.
  • CI/CD: at least 5 years as of 2026-09 (attested 2026-09-03). Covers: ci/cd.
  • Location: Jacksonville Beach, FL, US; remote available (attested 2026-09-03).
  • Work authorization: United States (attested 2026-09-03).
  • Work authorization: Canada (attested 2026-09-03).
  • Founder · Fincel Design, LLC · 2008–present
  • Founder · Geol.ai · 2025-08–present
  • AI infrastructure lead · B2B marketing agency (name withheld by policy) · 2018-10–present

Attested · 2026-09-04 · Kevin Fincel

These are the profile facts I've confirmed for Ask Kevin. It quotes them directly and does not infer additional profile facts.

§ Cookies

Cookie preferences

Choose what this site may keep in your browser. Details for every item are in the Cookie Policy. You can change this at any time from the bottom of any page. Cookie Policy.