How to Scope a Custom AI Model Project (and What to Charge)
The discovery questions, project types, price ranges, and scope management strategies for custom AI model projects. How to scope correctly before you quote anything.
Read the articleNotes from
the work.
Guides, research notes and what we learn building models.
Shopify AI Assistant Without OpenAI API Costs: The Local Model Approach
Shopify stores spending $500-5,000/month on AI API costs can replace those calls with a local fine-tuned model. Here's the architecture, the Shopify integration, and the cost math.
Read the article
The Solo AI Agency Tech Stack: 8 Tools, Zero Full-Time Hires
Running an AI agency solo in 2026 is possible with the right stack. Here are the 8 core tools, what each costs, and what they let you accomplish without hiring.
Read the article
How to Start an AI Automation Agency in 2026: The Full Playbook
The complete guide to starting an AI agency in 2026. Covers the three agency models, technical foundation, finding clients, pricing, and the 90-day launch plan.
Read the article
v0 App AI Features at Flat Cost — No Per-Token Pricing
v0 by Vercel makes AI features easy with the Vercel AI SDK. Here's how to replace per-token cloud API costs with a fine-tuned local model at flat monthly cost.
Read the article
Vibecoder AI Cost Guide: Every Major Builder Platform Covered (2026)
The complete AI cost guide for vibecoders using Bolt.new, Replit, Lovable, Cursor, Windsurf, v0, and Bubble. How each platform hits the API cost cliff and how to fix it.
Read the articleThe Vibecoder's Guide to Building an AI Moat (Not Another Wrapper)
Four types of AI moat, why prompts are not one of them, and the practical roadmap for vibecoders to build genuine technical defensibility with fine-tuned models.
Read the articleWindsurf Apps: Swap OpenAI for a Local Model
Windsurf writes OpenAI API calls by default. Fine-tune a model on your domain, serve it with Ollama, swap the base URL, and cut per-token cost to zero.
Read the articleCI/CD for Fine-Tuning Pipelines: Automating Train-Evaluate-Deploy
Manual fine-tuning doesn't scale. Learn how to build a complete CI/CD pipeline that automates training, evaluation, promotion gates, and deployment for fine-tuned models.
Read the articleThe Cost of Not Retraining: How Stale Models Quietly Break Production
Models degrade silently. A support bot trained on old docs, a classifier missing new categories, a client model that feels 'generic' — stale models cost more than retraining ever will.
Read the articleFine-Tuned Model Drift: Monitor and Retrain
Catch drift in a fine-tuned model before your users do. Three drift types, four detection methods, dashboard thresholds, and when a retrain is worth it.
Read the articleFine-Tuned Model Ops: The Complete Lifecycle Guide
The full lifecycle of fine-tuned models in production — from data preparation through deployment, monitoring, and retraining. Stage-by-stage breakdown with time estimates, maturity levels, and failure modes.
Read the articleFine-Tuning for AML Transaction Monitoring: Reducing False Positives
Banks spend $30B+ annually on AML compliance while rule-based systems generate 95%+ false positive rates. Learn how fine-tuning local models can cut false positives by 40-60% while maintaining 99%+ true positive capture — without sending transaction data to cloud APIs.
Read the articleHow Much Data Do You Actually Need to Fine-Tune?
What you get at 100, 1,000 and 10,000 training examples, where the returns curve flattens, and how to run a learning curve analysis on your own task.
Read the articleManaging 50+ LoRA Adapters in Production
Naming conventions, metadata sidecars, registries, and multi-LoRA serving for teams running dozens of adapters across clients, tasks, and base models.
Read the articleModel Risk Management for Fine-Tuned LLMs: SR 11-7 Compliance Guide
A practical guide to applying the Federal Reserve's SR 11-7 model risk management framework to fine-tuned LLMs in banking. Covers documentation requirements, validation frameworks, auditor questions, and why on-premise deployment simplifies compliance.
Read the articleOn-Premise AI for Banking: Satisfying Regulator Audit Requirements
Architecture and operational guide for deploying on-premise AI in banking environments that satisfy OCC, FINRA, and Federal Reserve audit requirements. Covers infrastructure, audit trails, access controls, change management, disaster recovery, and a 10-dimension compliance comparison.
Read the articleOn-Premise Healthcare AI: Architecture and Infrastructure Guide
A practical infrastructure guide for deploying AI on-premise in healthcare environments. Covers hardware requirements, network architecture, air-gapped deployment, HIPAA audit logging, model update strategies, and real cost comparisons against cloud APIs.
Read the articleFine-Tuned Model Rollback: Deployment Strategies
Blue-green, canary, and shadow deployment strategies for rolling a bad fine-tuned model back in seconds, plus the decision framework for when to pull it.
Read the articleSOC 2 and AI: Why Financial Firms Need On-Premise Model Deployment
Every AI API you add expands your SOC 2 audit scope. On-premise model deployment keeps AI capabilities within your existing security boundary — no new vendors, no new risk assessments, no scope creep. Here is how to deploy AI that your auditors will approve.
Read the articleThe Vibecoder's AI Stack: Lovable + n8n + Ertas + Ollama
The complete 2026 tech stack for builders who want AI-powered apps without per-token pricing. Build with Lovable, automate with n8n, fine-tune with Ertas, deploy with Ollama.
Read the articleAI Agency Opportunity in Financial Services: Compliance-First Positioning
Financial services firms spend more on compliance than any other industry. They need AI but can't use cloud APIs. Agencies that understand financial regulation have a $50B+ market opening. Here's your playbook.
Read the articleAI Agency Opportunity in Healthcare: Selling to Hospitals and Clinics
Healthcare AI spending is growing at 24% CAGR, but hospitals lack ML teams. Agencies that understand HIPAA compliance have a defensible moat. Here's the market, service packages, sales motion, and revenue model.
Read the articleThe AI Independence Checklist: 7 Signs You're Too Dependent on a Single Provider
A self-assessment checklist for AI vendor dependency. Score yourself on 7 warning signs — from single-provider concentration to prompt engineering as duct tape — and get actionable next steps for each risk level.
Read the article