John Meyer / Unity with AI

Applied AI, in projects and writing.

A portfolio and working notebook for retrieval, evaluation, guardrails, local models, runtime design, and the infrastructure choices that make AI useful outside the demo.

AI Without the Bullsh*t

Latest posts from the notebook.

Archive
Strategy 4 min

How I Think About AI for Small Businesses

Most small businesses don't have an AI problem. They have a workflow problem that AI might help with. Here's how I think about the difference — and what actually matters when scoping this kind of work.

Mar 16, 2026 Read →
Systems 4 min

The Difference Between a Demo and a System

Every AI vendor has a great demo. Six months after deployment, the reality is usually different. Here's what separates a demo from a system that actually runs in production.

Mar 9, 2026 Read →
Workflow 4 min

Your CRM Matters More Than Your Model

Business owners shopping for AI ask which model to use. That's the wrong question. The data layer — your CRM — is what actually determines whether AI works in your business.

Mar 2, 2026 Read →
Strategy 4 min

How to Evaluate an AI Vendor Without Getting Bullshitted

Every AI vendor demo looks great. The questions below are designed to find the gap between the demo and what actually happens after you sign.

Feb 23, 2026 Read →
Workflow 4 min

The 5 SMB Workflows Worth Automating First

Most businesses try to automate the wrong things first. Here are the five workflows where small and mid-sized businesses consistently see the highest return — and why order matters.

Feb 16, 2026 Read →
Systems 4 min

When AI Must Escalate to a Human

An AI that won't hand off a call isn't a good AI — it's a liability. Escalation isn't failure. Here's how to define it, design it, and build the handoff correctly.

Feb 9, 2026 Read →
Voice 4 min

What an AI Voice Agent Should Never Say

Most voice agent failures aren't technical — they're conversational. Specific phrases destroy caller trust immediately. Here's what to cut and what to say instead.

Feb 2, 2026 Read →
Strategy 4 min

The Hidden Cost of DIY AI

Building your own AI system looks cheaper than buying one. The upfront math usually works. The ongoing math rarely does. Here's what gets left out of the DIY estimate.

Jan 26, 2026 Read →
Strategy 4 min

Why Most AI Projects Fail Before the Model Matters

When an AI project fails, everyone blames the technology. Almost never the right diagnosis. Here are the four things that kill AI projects before the model gets a chance to help.

Jan 19, 2026 Read →
Voice 4 min

AI Voice Agents Explained in Plain English

Most people think AI voice agents are smarter phone trees. They're not. Here's how the technology actually works — and what has to be true for it to work in your business.

Jan 12, 2026 Read →
Workflow 4 min

What AI Actually Does for a Small Business

Most small business owners think AI means chatbots or job replacement. Here's what AI actually does in real operations — and why the boring use cases are the ones worth paying for.

Jan 5, 2026 Read →
01 - Projects

Current Project Briefs.

Active builds, structured notes, and the proof surfaces behind them.

Selected project briefs

Experiments, live systems, and tooling notes with the brief, the proof surface, and the implementation trail kept together.

tooling

Complete

RAG Drift Monitor

Semantic Drift Watcher

A monitoring layer that compares live retrieval output against stored intent baselines so drift shows up before the model starts sounding wrong.

Python sentence-transformers (all-MiniLM-L6-v2) ChromaDB Streamlit
Foundation Case Study →

tooling

Complete

Model Routing Middleware

Cost-Aware Prompt Router

A FastAPI middleware layer that inspects task shape and routes requests to the lowest-cost model that can still do the job well.

FastAPI Model routing Cost telemetry Threshold-based fallback
Foundation Case Study →

experiment

Complete

Recursive Agent Loop

Patch-and-Retry Sandbox

A contained patch-and-retry loop where an agent observes its own failures, proposes changes, and keeps iterating until the test passes or the budget runs out.

Sandboxed execution Traceback parsing Patch loop orchestration Test gating
Foundation Case Study →

prototype

Complete

Hybrid Retrieval Pipeline

BM25 + Dense + Reranking

A retrieval pipeline that blends BM25, dense vectors, and a cross-encoder reranker so exact-match precision and semantic recall can work together instead of competing.

BM25 (Okapi, implemented from scratch) sentence-transformers (all-MiniLM-L6-v2) Cross-encoder reranking (ms-marco-MiniLM-L-6-v2) Reciprocal Rank Fusion (RRF)
Frontier Case Study →

prototype

Complete

PII Redaction Proxy

Outbound PII Filter

A proxy layer that runs a regex pass for structured PII and a BERT-NER pass for unstructured entities, combining both before any outbound model call.

Regex filtering (8 structured entity patterns) BERT NER (dslim/bert-base-NER, CoNLL-2003) Two-pass detection with span deduplication Context preservation measurement
Frontier Case Study →

live system

Live Demo + Iteration

Steve — Voice Agent Sandbox

Public Runtime + Recap Pipeline

A public voice-agent project with signed access, recap lookup, transcript handling, rate limiting, and a private regression harness behind the scenes.

ElevenLabs Astro Cloudflare Signed session URLs
Live System Case Study →

experiment

Research + Prototype

Gemma on Metal

Local Gemma 4 + LoRA on Apple Silicon

A practical test of Gemma 4 on a 16GB MacBook M2: run the small edge model locally with Ollama and Metal, then shape a LoRA tuning path for private document and audio extraction.

Gemma 4 Ollama Apple Silicon Metal
Local AI Case Study →
02 - How I Work

What I Focus On.

Three things show up in every project: making the runtime work, keeping the boundaries secure, and measuring whether the system is actually improving.

[ APPLIED AI ]

Agentic Systems

  • Model routing middleware
  • Eval + drift detection
  • RAG observability

[ CYBER ]

Security-First AI

  • PII redaction proxy
  • Guardrails + abuse resistance
  • Trust boundary design

[ CLOUD ]

Platform Reliability

  • Token cost control
  • Runtime telemetry
  • Failure boundary engineering

The projects show these patterns in action. The writing archive explains the decisions behind them without forcing everything into one long homepage narrative.

03 - About

How I Got Here.

John Meyer

I work in security engineering and spend a lot of my time thinking about how systems behave under pressure. That perspective carries directly into the AI work here: clean boundaries, clear observability, and less trust in demo-only assumptions.

Unity with AI is the place where that background turns into projects and writing. It is less a storefront and more a durable record of the kinds of systems I want to keep building.

LinkedIn Writing Archive
20+ Years
Army IT to Agentic AI
Security Engineer
Identity, Endpoint, Detection
SaaS Builder
Co-Founder, Cloud, and Ops
U.S. Army
IT Specialist
04 - Connect

Get in Touch.

If something here overlaps with what you are building, send a note. A thoughtful message is more useful than a perfect intake form.

Good reasons to reach out

  • You are hiring for applied AI, platform, or security-minded systems work.
  • You want a second set of eyes on an architecture, retrieval stack, or agent workflow.
  • You just want to compare notes on a hard build, a weird failure mode, or a direction you are exploring.

Email is enough. Phone is optional.

COLLABORATION FORM · UNITY WITH AI