TAG

Zack Martin · Portfolio

Applied AI, built by someone who has run the operation.

I'm the first Applied AI Solutions Engineer at Solid, a seed-stage startup. I led the shaping of its solutions engineering program and tooling, and I engineer the semantic models and AI agents that enterprise customers put into production. I got here through 15+ years at Best Buy: field service technician, then running delivery and installation logistics across five states, then supply chain transformation.

Outside my full-time job, I chair the board of Wild Coast Trails Association, a trail nonprofit. I also run Tactical Adaptability Group, a freelance consulting practice that teaches AI skills and helps organizations automate their work, and I built TrailAdvocate, a SaaS platform for trail organizations.

Full-time career
Best Buy (15+ years, enterprise retail) → Solid (2024–present, seed-stage AI startup)
Side projects
Wild Coast Trails Association (board chair) · Tactical Adaptability Group · TrailAdvocate · Silicon Marsh Ventures
Zack Martin

Results at a glance

Career progression and side projects

Best Buy and Solid are my full-time career, from field service technician to applied AI. My side projects started in 2022 and run alongside it.

Full-time career milestones

  1. Solid · 2025–nowApplied AI Solutions EngineerFirst Solutions Engineer: led the shaping of the program and tooling, and took pilots to productionRead the story ↓
  2. Solid · 2024–2025Product Manager, Analytics CommunityA growth role I defined myself, with near-total autonomyRead the story ↓
  3. Best Buy · 2022–2024Senior Analyst, Strategy & TransformationLed C-suite initiatives across the national DC networkRead the story ↓
  4. Best Buy · 2021–2022Analyst, Vendor SupportTaught myself EDI to automate a major vendor's order-status updatesRead the story ↓
  5. Best Buy · 2018–20213PL Regional ManagerDelivery and installation for 5 states and 28 storesRead the story ↓
  6. Best Buy · 2008–2018Field Service TechnicianMECP Master certified (1 of 270 nationwide) and market trainer for new techniciansRead the story ↓

Between my two Best Buy stints, from October 2015 to September 2016, I was a product sourcing and testing specialist at Super Bright LEDs.

Side projects · since 2022

Hiring for a specific role? Filter the stories by focus.

Solid · 2024–present · full-time · seed-stage startup

Applied AI at Solid

Solid automates creating, maintaining, and testing semantic models, using a company's warehouse, BI tools, and query history, and serves those models to AI agents over APIs and MCP. As its first Solutions Engineer, I engineer semantic models for enterprise customers and connect them to the agents that do the work.
Full-time · SolidApplied AI Solutions Engineer (Solid's first)Dec 2025 – present

Took Solid's first enterprise customer from pilot to AI agents in production

When I started in the role, Solid had no enterprise customers. A major enterprise SaaS company was testing basic chat-with-your-data tools that weren't answering business questions reliably. I built a plan, with use cases and a pilot timeline, to give them semantic models their agents could rely on to do real work. The pilot became Solid's first recurring subscription, and the customer's agents are now in production.

What I did

  1. Scoped the engagement as a business-validation pilot. I wrote the use cases, success criteria, and timeline, and got sign-off from the customer's stakeholders.
  2. Created Solid's intake and spec process for pilots, with templates for use cases, security review, and what each side needs to deliver.
  3. Engineered the customer's semantic models and ran structured evaluations and benchmark rounds until the customer certified them at 94–100% accuracy.
  4. Built custom scripts and agent templates for the customer, then supported their team hands-on through the production push.
  5. Documented the entire Solid product on a public docs site and trained the customer's team on semantic modeling.
  6. As Solid's first Solutions Engineer, led the shaping of its solutions engineering program and tooling: a solutions library, reference architectures, and shared repositories for evaluation assets.
  7. Shaped many of the major features in Solid's product today through specs I wrote from customer work.

  • 1st

    Enterprise customer and recurring subscription in Solid's history

  • 94–100%

    Certification accuracy I consistently reach on the semantic models I engineer

  • In production

    The customer's AI agents, running on models I built

How I run a pilot
  1. Discovery
  2. Written spec
  3. Build & integrate
  4. Validate
  5. Certify
  6. Enable
More detail

Tools & tech

  • Semantic modeling
  • Structured evaluations
  • Benchmark question sets
  • Python
  • SQL
Full-time · SolidApplied AI Solutions EngineerDec 2025 – present

Building AI agents that answer correctly and take action

Any LLM can write SQL. Very few know what "active customer" means at your company. Solid's semantic layer carries that meaning; my work is wiring it into the agent tools enterprises already run, and fixing models when the answers drift.

What I did

  1. Connected Solid to the agent frameworks enterprises already use, including CrewAI, Langflow, Microsoft Copilot Studio, Workato, Claude, OpenAI, and MCP-based agents, so agents reach data through governed definitions instead of raw tables.
  2. Built agents that get governed SQL from Solid, run it against the warehouse, and act on the results: posting to Slack, taking actions in business systems, and recommending next steps.
  3. Built a standard agent framework and applied it to use cases in health-insurance claims, pharmacy benefits, banking, and hospitality operations. The agents can take actions across many different systems, with a human in the loop, and work the same way connected to a customer's environment. I share the code with customers who are new to building agents, so their teams can replicate it.
  4. Built a semantic-model improvement agent from what I learned on a stalled pilot. It builds models, checks question coverage, proposes fixes, and flags instruction issues. I architected it to run first as an internal back-office tool and later inside Solid's product, to improve customer outcomes.

  • Any framework

    CrewAI, Langflow, Copilot Studio, Workato, Claude, OpenAI, and MCP-based agents

  • Human in the loop

    Agents that take action across business systems, with a person approving the consequential steps

Where Solid sits

Context and data sources

Data assets
Snowflake, Databricks, BigQuery
Documentation
Confluence, Notion, SharePoint
Context
Slack, Teams
Metric definitions
Tableau, Power BI, Looker

Solid semantic layer

Know-how, glossary, and metadata
Semantic models
Generation engine
SQL, metadata, and chat
Solid MCP

AI agents and delivery

AI agents, any framework
CrewAI, Copilot Studio, Claude, OpenAI, Langflow, Workato
Slack and web users

One question, end to end

  1. 1. Someone asks a question in Slack or on the web
  2. 2. Solid MCP generates SQL from the semantic model
  3. 3. The SQL runs in the warehouse with that user's permissions
  4. 4. The agent analyzes the results and acts
  5. 5. Insights, raw data, and the query go back to the user
Human in the loop: agents investigate, recommend, and act across systems, and a person approves the consequential actions.
Adapted from the Architecture Configurator I built for Solid. Solid reads metadata, never moves the data, and generates the SQL; it runs in the customer's warehouse under their own permissions.
More detail
  • Also contributed to the Apache Ossie project by writing a converter for Solid's semantic models: a Python package that converts models between Solid and Ossie in both directions, offline.
1. Diagnose
Which questions fail, and why?
2. Choose a lever
Coverage, instructions, or model fix
3. Apply
One change at a time
4. Re-run
Benchmark again

Repeat until the benchmark moves

Before changing anything, ask which layer failed: the semantic model, the platform, the calling agent, or the evaluation itself.
The loop the semantic-model improvement agent runs.

Tools & tech

  • MCP
  • CrewAI
  • Langflow
  • Copilot Studio
  • Workato
  • Claude
  • OpenAI
  • Python
  • Snowflake
  • Cursor

What I learned

The first question is always which layer failed: the semantic model, the platform, the calling agent, or the evaluation. Changing a model to fix what is really an agent problem just moves the failure somewhere else.
Best Buy · 2008–2024 · full-time

Operations and supply chain at Best Buy

I spent 15+ years at Best Buy, starting as a field service technician and finishing on the supply chain strategy and transformation team. Along the way I learned that a new system only matters once the people in the field use it.
Full-time · Best BuyAnalyst, Supply Chain Vendor SupportAug 2021 – Jul 2022

Replacing a phone-and-email process that lost inventory

When a customer's special order was stuck, the stock often sat in a distribution center that didn't share a shuttle with the one fulfilling the order. Moving it took a manual out-of-network transfer arranged by phone and email, with no tracking: transfers got auto-cancelled or lost in carriers' LTL networks, and nobody knew what the process cost.

What I did

  1. Talked to the operators and subject-matter experts first, to confirm automation was worth it.
  2. Pulled together a cross-functional committee and designed the target flow on a Miro board.
  3. Built a SharePoint list with custom forms that capture only what matters: product and transport cost, origin and destination, supplier, transfer number.
  4. Wrote a Power Automate flow keyed to each transfer status: request a carrier bid, notify the sourcing DC, track the shipment, and tell the requester when to schedule the customer.
  5. Ran UAT with test transfers, launched on a national go-live call with the DC network, and held bi-weekly support and feedback calls through the change.

  • ~$650K

    Inventory moved through the new process

  • 1 week

    Faster transfer execution

  • 1.5 hrs

    Labor saved per transfer, end to end

The automated transfer, status by status
  1. Transfer created
  2. Carrier bid
  3. Sourcing DC notified
  4. In transit, tracked
  5. Received
  6. Customer scheduled
More detail

Tools & tech

  • SharePoint Lists
  • Power Automate
  • Miro
  • Reporting
Full-time · Best BuyAnalyst, Supply Chain Vendor SupportAug 2021 – Jul 2022

I taught myself EDI to fix a vendor's order-update gap

Customers on backorder weren't hearing anything. A major appliance vendor was sending about 250 purchase-order updates a month against an order pool of more than 10,000, and didn't have the engineering capacity to fix it on their side.

What I did

  1. Built the business case for fixing it on our side and got it approved.
  2. Mapped which reporting sources showed what needed an update, and when.
  3. Learned the EDI 855 spec and the related documents in the CommerceHub integration, then wrote a Python job that turns an exported CSV into a valid EDI feed.
  4. Worked through the risk of updating live customer orders with cross-functional partners, then tested end to end with the vendor's EDI engineers on dummy POs.
  5. Shipped it, monitored it, and documented it so a future owner, or the vendor, could run it.

  • 80% → 3%

    Late backorders with no customer communication, within six months

  • 500–1,000

    Orders updated twice a week, up from ~250 a month

Late backorders with no customer update
Before80%
Six months later3%
More detail
  • In the same role, the vendor inventory strategies I led cut out-of-stock backorders from 40% to 8%, and my sales and operations planning (S&OP) support lifted customer satisfaction on delivery dates by 15%.

Tools & tech

  • Python
  • EDI (855 and related documents)
  • CommerceHub
  • CSV exports
  • S&OP
Volunteer and founder work · 2022–present

Side projects

Each of these started with a problem close to home: a volunteer trail club that had outgrown its structure, trail work tracked in spreadsheets and group texts, and local organizations asking how to use AI well. I've taken each one from an idea to something people rely on.
Founder · side projectLiveFounderMay 2025 – present

Built software that solved a real problem for my trail stewardship group

Running a trail nonprofit meant tracking conditions, volunteer work, and grant reporting across dozens of trails in spreadsheets and group texts. I built the tool I wanted, then turned it into a business.

What I did

  1. Built and launched the product end to end: automated trail condition updates, work and issue tracking, and grant-ready reporting.
  2. Added a volunteer sign-up and waiver tracker that automates event registration and hour logging.
  3. Commercialized it with pricing based on organization size, and signed paying customers.
  4. Self-hosted the application, with automated DevOps workflows, monitoring and alerting, and automated tests.
  5. Built a team of AI agents that handles marketing, DevOps, and monitoring tasks for the business.
  6. Formalized it as TrailAdvocate LLC in June 2026 and brought on a business partner.

  • Live

    SaaS at TrailAdvocate.app

  • Paying customers

    Commercialized, with paid plans for larger organizations

More detail
TrailAdvocate.app
Founder · side projectFounder & Principal ConsultantOct 2024 – present

Teaching the coast to use AI, and building the platform to do it

Through Tactical Adaptability Group I help small businesses, nonprofits, and public organizations in Southwest Oregon put AI to work responsibly. Scaffold, the AI learning platform I built, is where that teaching scales.

What I did

  1. Taught "Level Up with AI: Smarter Tools for Tourism" at the Oregon Coast Visitors Association's 2025 People's Coast Summit, plus workshops for Siuslaw Vision, the Bay Area Chamber of Commerce, and the TSOC Network Conference.
  2. Took on paid work in grant writing for Siuslaw Vision and professional development for OSU, and contracted an AI professional-development series for Lane ESD career and technical education (delivery pending).
  3. Built Scaffold solo: courses on LLMs, RAG, and agents, an AI tutor with pgvector retrieval and golden-set evals, a project builder, and Stripe billing, on TanStack Start, Node, and Postgres.
  4. Built it with an agentic engineering workflow: AI coding agents working under written guardrails and my review.

  • ~10

    Pro bono AI workshops in two years

  • ~120

    Attendees at my TSOC Network Conference workshop

More detail
Speaking record →Services →Testimonials →scaffold.tacticaladapt.com

Other projects

Tools and integrations that didn't need a full story.

  • Starburst integration, end to end

    Solid · Sep 2026

    Contributed to Solid's Starburst integration and demonstrated the full workflow: integration and asset documentation, model building, and export to Starburst Data Products.

    A shared team effort, working end to end in a development release.

  • Solid ↔ Apache Ossie model converter

    Solid · 2026

    Contributed to the Apache Ossie project by writing a converter for Solid's semantic models: a self-contained Python package that converts models between Solid and Ossie in both directions, offline, with no Solid connection required.

How I run a pilot or a rollout

Pick a step to see how I've run pilots and rollouts in different organizations and roles.

Step 1 · Discover

Understand the workflow, the people in it, and the decision the work should change.

At Best Buy · supply chain

I talked to operators and subject-matter experts before automating inter-DC transfers, to understand the problem and its impact.

At Solid · applied AI

I run full customer discoveries before proposing a pilot, to understand what it costs the customer when they can't trust their AI agents' answers.

Step 2 · Write the spec

Turn wish lists into a written scope and success metrics everyone agrees to.

At Best Buy · supply chain

I defined exactly what the field needed to see in yard reporting (status, contents, and age) before anything was built.

At Solid · applied AI

I built templates for defining pilot use cases, the security review, what we need from the customer and from our own team, and internal specs. I use them on every pilot, and so does my team.

Lesson learned: Write the scope and success metric down before kickoff, not after the first disagreement.

Step 3 · Build & integrate

Build it, or partner with the team that does, on whatever technology the organization uses.

At Best Buy · supply chain

I mapped transportation and warehouse system tables in SQL, wrote the EDI feed in Python, and built transfer flows in Power Automate.

At Solid · applied AI

I engineer the semantic layer, then connect it to the customer's agents through MCP, whatever agent tools they plan to use.

Lesson learned: When something fails, do a root-cause analysis. Find which layer failed before changing anything.

Step 4 · Validate & certify

Test with real users and real questions, revise, and get explicit sign-off.

At Best Buy · supply chain

I ran a two-week user acceptance test with field users and transportation analysts, then revised the reports with decision science until sign-off.

At Solid · applied AI

I build the models, then run structured evaluations and benchmark rounds, adding question coverage until the models can be certified at 94–100% accuracy.

Step 5 · Enable & roll out

Training, documentation, and support until people are comfortable with the solution and adopt it.

At Best Buy · supply chain

I recorded training videos, wrote job aids, and held live training sessions on national DC network calls, then ran bi-weekly office hours after go-live.

At Solid · applied AI

I ran training sessions, built a public documentation library, and supported the customer hands-on through their production push.

Lesson learned: Adoption takes support after launch, so I plan for it from the start.

Step 6 · Handoff and feedback

Give the result a clear owner, and turn what users learned into what gets built next.

At Best Buy · supply chain

After testing, I handed ownership of the yard report to the field, so field leaders run it and are accountable for the number themselves.

At Solid · applied AI

I turn what I see in customer work into feedback and specs for Product and R&D, covering model-building usability, benchmarking workflows, and observability.

Skills and tools

Product & AI adoption

  • Technical product management
  • Roadmaps
  • AI change management
  • Stakeholder alignment
  • Success metrics & KPIs

Agentic systems

  • Semantic modeling
  • Agentic workflows
  • MCP integrations
  • Evaluation & benchmarking
  • Pilot-to-production delivery

Engineering & tools

  • Python
  • SQL
  • REST APIs
  • EDI
  • React
  • TypeScript
  • CrewAI
  • Langflow
  • Copilot Studio
  • Workato
  • Power Automate
  • Cursor
  • Claude
  • OpenAI
  • Streamlit
  • Power BI

Operations & analytics

  • Supply chain transformation
  • TMS / WMS / ERP
  • Cross-functional leadership
  • UAT & change management
  • Lean Six Sigma (Green Belt)
  • Tableau

Let's talk

Building a product, applied AI, or architecture team?

I'm always glad to share ideas and talk with teams working on enterprise AI adoption. Email me or connect on LinkedIn.

Need help with AI in your organization?

Through Tactical Adaptability Group, I run workshops and hands-on consulting for small and medium-sized businesses, nonprofits, and public organizations, on-site in Southwest Oregon or remotely anywhere.

Work at Solid is described with customer names withheld; links go only to public assets. Based in North Bend, Oregon.