James Pritchard
Architect of enterprise AI infrastructure and generative asset pipelines. Specialist in building deterministic wrappers and micro-prompt orchestrated systems that treat LLMs as semantic components within reliable Go and TypeScript environments. Expert in Behavioral AI Engineering — leveraging psychographic data and multi-stage research pipelines to ground probabilistic models in verifiable truth.
Technical Skills
{
"ai_infrastructure": [
"Provider-Agnostic Rate Limiting",
"Multi-Pass Validation",
"Semantic Classification"
],
"generative_engineering": [
"Micro-Prompt Orchestration",
"Psychographic Targeting",
"GEO/SEO Synthesis"
],
"systems_engineering": [
"High-throughput Model-Specific Queuing",
"Token Reconciliation",
"Linux Internals"
],
"the_stack": [
"Go", "TypeScript", "tRPC", "React", "HTMX",
"MongoDB", "Redis", "GCP", "AWS"
]
}Experience
Lead AI Engineer — Enrollment Resources · Mar 2024–Present
Unified AI Template Architecture: Engineered a core architectural layer that powers a suite of generative products. Designed a Micro-Prompt Orchestrator that utilizes a "Constraint Hierarchy" (XML-tagged data, banned-phrase lists, and exact sentence-count gates) to eliminate "AI slop" and ensure 100% brand integrity.
Psychographic Personalization Engine: Developed a targeting system that maps prospect form data (Social Styles: Expressive/Amiable/Driver/Analytical) directly to LLM system prompts. Dynamically shifts vocabulary and motivations based on statistical distributions of actual student behaviors.
Deep Research & GEO Pipeline: Architected a deterministic 11-step research tool that performs iterative query refinement and source-quality scoring (Relevance, Credibility, Quality, Recency). Implemented a Competitor Exclusion Filter to suppress rival .edu domains while boosting neutral authorities (BLS, StatCan) to maximize search and discovery engine (GEO) rankings.
Provider-Agnostic AI Rate Limiter (Go/Redis): Designed a high-performance Go microservice to coordinate global AI API usage. Engineered model-specific queues (Sonnet, Gemini Pro/Flash, etc.) utilizing a Predictive Token Gatekeeper to maintain a consistent 95% utilization of quotas while eliminating 429 rate-limit resets.
Two-Pass Quality Gate: Developed a tiered model strategy (Gemini Medium/Low Thinking) that performs real-time schema validation on content blocks, executing automated, surgical repairs for failed components.
Visual Asset Pipeline: Built an image generation engine with Field-Accuracy Guards and two-step logo integration (using Gemini Vision). Engineered "Brand Color as Physical Object" logic to naturally incorporate hex codes into generated scenes.
Platform Modernization: Executed a high-velocity architectural migration of the "Virtual Adviser" platform (v6) from legacy Node/Express to a modern, type-safe TypeScript and tRPC stack.
Chief Technology Officer — VirtuOHS · 2024–Present
Deterministic Agentic Workflows: Developed a safety-critical hazard assessment platform using Go and HTMX. Utilizes LLMs strictly for semantic classification while maintaining all mathematical risk-modeling logic (William Fine Model) in reliable Go code.
Human-in-the-Loop Orchestration: Designed a confirmation-driven state machine that requires user verification of extracted data points before triggering secondary report-generation agents.
Junior Developer — Enrollment Resources · Jan 2023–Mar 2024
Asset & Site Management: Onboarded clients and resolved support tickets while developing "assets" (forms, quizzes, landing pages) using the Virtual Adviser Editor (block-based builder). Managed site migrations across WordPress, Wix, and Hugo using Digital Ocean.
Maintenance: Handled core bug fixes and general development for the Virtual Adviser v5 platform.
Freelance Web Developer · 2022–2023
Built an inventory management app for a local coffee roaster, consolidated databases for a non-profit, and wrote automation scripts for a telecom installer. Small scopes, real constraints.
System Technician — Shaw Cable Systems Ltd. · 2019–2021
Maintained infrastructure uptime for 16,000+ customers and led network upgrades increasing bandwidth from 600 Mbps to 1.5 Gbps.
Selected Projects
Suede — A Programming Language for AI Workflows
A programming language for AI workflows where cost is visible in the syntax. Every line that calls a model is marked with ~=. Every line that doesn't is plain code. You can read the price of a workflow by scanning for the squiggles.
Six built-in model verbs (extract, classify, compress, rewrite, expand, generate) with baked-in prompt templates. Typed input goes in, typed JSON comes out. The model does one job and sits back down.
The static analyzer estimates the cheapest and most expensive paths through a program before you spend a cent. Zero-config rate limiting learns your provider's limits from 429 responses and queues around them automatically. Mix providers (OpenAI, Anthropic, Gemini) in the same program.
Written in pure JavaScript. Runs as a CLI, Node library, or entirely in the browser. The playground is the full interpreter running client-side.
MENACE — Provider-Agnostic AI Coding Command Center
A command center for AI-driven code changes that separates planning from execution. An Architect agent plans changes using purpose-built navigation tools (symbol lookup, call graphs, AST traversal via ctags), produces a reviewable proposal, and dispatches subtasks to parallel Worker models with conflict-aware scheduling. The custom navigation tooling, benchmarked against standard Read/Grep/Glob on MENACE's own codebase, uses 33x fewer tokens to locate and read a specific function and 108x fewer tokens to assess change blast radius. Provider- and model-agnostic (Anthropic, Google, OpenAI, local Ollama); features Vim keybindings, a Lua-extensible tool system, SQLite session persistence, and OS keychain auth.
Writing
Author of james-pritchard.com/blog — essays on building reliable LLM features for production rather than demos, covering AI code review, human-in-the-loop design, agentic-first architecture, and treating LLMs as typed functions.