# Kyle Fahey

> Product and engineering leader, Phoenix, Arizona. Start with what’s true. Build what follows. Four years in Medellín, and months across Latin America and Iberia: I stay until I can tell what holds everywhere from what’s only local. I build the same way. Lately that means AI products where being wrong is expensive: deciding what they should do, then building them. And writing about what AI will actually change. Before that, nine years building for AbbVie (Allē), OC Sports & Entertainment (Anaheim Ducks), a Y Combinator startup, and Thought Catalog.

Contact: the form at https://kylecrfahey.com/#contact. Links: LinkedIn https://www.linkedin.com/in/kylecrfahey/, GitHub https://github.com/quesurifn, Medium https://medium.com/@kyle.c.r.fahey.

## Work
- GARDE1: Compliance that does the work instead of handing it to you. CMMC readiness for defense manufacturers with no security team. I built the product, the brand and the engineering. (https://kylecrfahey.com/work/garde1)
- PWTL (Prompt Workflow & Task Language): AI decisions that cite their evidence. A task language that asks AI to choose an answer instead of writing one, and to show its work. (https://kylecrfahey.com/work/pwtl)
- The GARDE1 document system: Documents that keep a receipt for every sentence. Compliance policies where every sentence traces back to the fact that required it. (https://kylecrfahey.com/work/documents)
- Genomica: Psychiatry’s first prescription is a guess. Genomica makes it a decision. Only about one in four people with depression reach remission on their first antidepressant. We’re building genetic clinical decision support for psychiatrists, mapping genome to mechanism to drug. (https://kylecrfahey.com/work/genomica)
- Seven Tests for Consciousness: Humans defined consciousness, so any test for it has to start from the human experience. Seven binary, independently testable criteria, each derived from what being conscious is like for us, for whether any system, biological or artificial, is conscious. (https://kylecrfahey.com/writing#papers)

## Inventions
- PWTL (Prompt Workflow & Task Language): A task language that makes AI judgments auditable. Patent pending.
- GARDE1 document system: Compliance documents that can explain why every statement in them is there. Patent pending.
- Locus: A protocol that gives AI a map of a website, so it can show non-technical users where things are by chat, without navigating them away. Protocol.

## Consulting offers (https://kylecrfahey.com/consulting)
- Architecture review. For: CTOs whose AI works in the demo and wobbles in production. You get: A written report and a ranked plan. Typical length: Two weeks.. Proof: Audited part of a $6M PwC engagement.
- An AI product, from zero. For: Founders who need one person to take an AI product from idea to launch. You get: Thesis, design, brand and a production system. Typical length: Eight to sixteen weeks.. Proof: GARDE1, built end to end.
- Turnaround and fractional CTO. For: Seed to Series B teams. You get: A stable platform and a team that can run it. Typical length: Three to six months, part-time.. Proof: Pangea (YC W21).
- Enterprise and compliance readiness. For: Startups selling into enterprise or defense. You get: SOC 2 or CMMC groundwork. Typical length: Four to eight weeks.. Proof: SOC 2 at Litehouse; GARDE1.
- Data and AI pipelines. For: Businesses with scattered data. You get: Pipelines and search. Typical length: Four to ten weeks.. Proof: LabRecruit.

## Case studies (https://kylecrfahey.com/work/case-studies)
### Steadying a YC startup’s platform, Pangea.app (YC W21), 2021–2022
Problem: A fast-growing marketplace with a non-technical founding CTO. The app leaked memory, crashed periodically, and slowed down under load.
- Advised the CTO directly on architecture, infrastructure and engineering practice.
- Replaced WebSockets that were being used like a REST API with real HTTP, which stopped the memory leaks.
- Moved slow third-party calls out of request handlers and onto queues.
- Migrated off unsupported RethinkDB to Postgres, ending the crashes.
- Led the move to TypeScript, linting and tracing (New Relic, Sentry).
Result: A stable, faster platform, fewer bugs shipped, and a CTO who could run it alone afterwards.

### Finding experts from what they’ve built, LabRecruit, London, 2019
Problem: A recruiting firm specializing in functional programmers needed to find rare engineers who weren’t on job boards.
- Found engineers through public GitHub commits in the target languages.
- Validated contact details and matched each engineer to their professional profile.
- Indexed the results in Elasticsearch so recruiters could search them.
Result: Recruiters could search by what someone had actually built, not by self-reported keywords. It’s the same identity matching and enrichment today’s AI recruiting tools are built on.

### Search for 2 million visitors a month, AbbVie (Allē), 2020–2022
Problem: A new booking and rewards platform for cosmetic treatment providers had no real search, and a federated GraphQL schema with no shared standards.
- Convinced leadership to adopt Elasticsearch instead of Postgres: provider taxonomies were bigger than Postgres could search comfortably. Introduced it to the organization and built its first search service.
- Architected the core Provider service in Go, with geolocation, email automation and secure image handling.
- Co-founded the GraphQL Council, which reviews every schema change across 20+ services. Federation at that scale needs decision makers, and they should be democratic, so the best experience is at the table.
Result: The service became the primary search on alle.com, with over 2 million monthly visitors.

### Accurate natural language to execution, Litehouse, 2024–present
Problem: Operators of a smart district, proven at the Honda Center, needed answers from a federated data platform without writing queries.
- Built RAG pipelines that turn questions like “How many tickets were sold last game?” into valid GraphQL queries.
- Built semantic search (vectors plus reranking) and a Lucene query builder over platform data.
- Put a Go proxy in front of the LLM tooling to enforce the platform’s attribute-based access control.
Result: Plain-language answers that respect who is allowed to see what.

## Experience (https://kylecrfahey.com/resume)
### Founder, GARDE1, 2025 – present
- Built the company’s product end to end: thesis, brand, design, marketing site and engineering.
- 8,731 commits across a C#/.NET endpoint agent, a Go backend, and a TypeScript frontend and connector runtime.
- Invented PWTL, a task language for auditable AI judgments, and a document system that explains every statement it generates (both patent pending), plus Locus, a protocol that gives AI a map of the product so it can guide non-technical users by chat.
- Set the strategy from a competitive analysis of Vanta, Delve and Drata; turned 27+ navigation items into the states of a GRC.

### Staff Product Engineer, Litehouse, 2024 – present
- One of five engineers chosen from thirty-three to turn a smart-district platform, proven at the Honda Center, into a commercial product.
- Owned backend architecture for the commercialization effort: designed the multi-tenant system, redesigned 40+ domain relationships, and architected the federated GraphQL service across our backends, spanning both the data warehouse and application data, to turn legacy trial systems into a scalable product platform.
- Designed and led the platform-wide Auth0 migration, modernizing authentication and authorization across services for security, maintainability and enterprise readiness.
- Developed and productionized RAG pipelines (vector and semantic search, OpenAI models, Python in Azure Synapse, TypeScript and Node.js) for natural-language querying over GraphQL-backed platform data; after auditing PwC’s pipeline, rebuilt it with a temporal engine, a trimmed schema and a skills-based semantic layer in an agentic loop.
- Invented an in-house framework on top of NestJS that adds MSSQL tables to the platform mechanically, and built the patterns and developer tooling around it to standardize backend development and speed delivery across distributed product teams.
- Audited part of PwC’s $6M platform delivery before we accepted it, while serving as the primary backend and GitHub Actions CI/CD expert, mentoring engineers and improving platform stability, delivery quality and SOC 2 readiness.

### Senior Product Engineer, OC Sports & Entertainment (Anaheim Ducks), 2022 – 2024
- Owned the backend TypeScript and NestJS codebase and set its quality bar: authored the house code guide and reviewed every pull request (1,000+), refusing to merge until it met the bar.
- Mentored an offshore team of 30 engineers in India through technical 1:1s.
- Audited Aegis, an open-source backend module federation framework, and wrote its stability and developer-experience roadmap; when it wouldn’t be ready for the Ducks’ new fan-experience app, pivoted to NestJS, trained 30+ engineers across the US and India, and shipped on time.
- Designed the backend architecture and the real-time game-notifications pipeline on Azure Event Hub, working through live-scoring data integrity.
- Designed authentication and authorization (ABAC, JWT) with integrations including Ticketmaster; led adoption of Datadog and GitHub Actions.

### Senior Software Engineer, AbbVie (Allē), 2020 – 2022
- Convinced leadership to adopt Elasticsearch over Postgres for provider search, and built the search service behind alle.com, with 2M+ monthly visitors.
- Architected the core Provider service in Go (Postgres and PostGIS, gRPC, GraphQL) within a federated graph of 20+ services.
- Helped voice the need for, then co-founded, the GraphQL Council, which sets schema standards and reviews every change across the graph.
- Taught Go to 60+ engineers through talks, workshops and written guides, and made the case for an opinionated language: a large team ships faster when the code is predictable.

### Independent consultant, advisor to the CTO, Pangea.app (YC W21), 2021 – 2022
- Mentored the non-technical founding CTO on architecture, infrastructure and practice.
- Took the platform from several crashes a day to 99.9% uptime: WebSockets to HTTP, slow calls onto queues, RethinkDB to Postgres.
- Made tests and pull-request reviews how the team worked.
- Led the move to TypeScript, linting and tracing, cutting bugs shipped.

### Senior Software Engineer, Hoopla, 2020
- Led a new Rails microservice publishing notifications to Microsoft Teams.
- Added a “fan” role across 13 microservices so colleagues could celebrate sales wins.

### Software Engineer, The Thought & Expression Co. (Thought Catalog), 2017 – 2020
- Helped sustain growth as revenue rose 300% in my first year, while improving uptime and performance; 300+ production deployments.
- Cut AWS costs 40% with no downtime; re-indexed MySQL for an 85% speedup; made Laravel controllers 60% faster.
- Built recurring payments (Stripe, Authorize.net, ACH), an Elasticsearch recommendation engine and MongoDB disaster recovery.
- Paired Elasticsearch with Ahrefs data to match existing content to trends; built a Slack bot that ran editorial workflows.

### Independent consultant, LabRecruit, 2019
- Built a pipeline that found specialist engineers from public GitHub activity, verified contact details, matched professional profiles and indexed them for search. LabRecruit later sold it to another agency.

### Full-stack developer (contract), Elexa Consumer Products, 2017
- Designed and built an e-commerce checkout customers called “the easiest checkout ever,” on a headless CMS with a Vue cart.

## Education
- Western Governors University, B.S. Computer Science
- Georgia Institute of Technology (OMSCS), graduate coursework in Computer Science

## Papers (https://kylecrfahey.com/writing#papers)
- Seven Tests for Consciousness: A Practical Framework for Evaluating Conscious Systems (Published, 2026). Existing frameworks for evaluating consciousness, including IIT, GWT and HOT, are hard to test or incomplete. This paper proposes seven binary, independently testable criteria: open system, systemic state, irreproducibility, continuous capacity, multi-modal learning, integrated and integral memory, and superposition. It predicts that classical binary hardware cannot meet the conditions for consciousness, that bees qualify while jellyfish do not, and that AGI requires substrates capable of genuine superposition.
- Why Today’s AI Is Not Conscious: Applying the Seven Tests (In progress). The follow-up: the seven tests applied to large language models, one at a time, with the reasons each one fails.

## Essays (https://kylecrfahey.com/writing)


## Side projects (https://kylecrfahey.com/workbench)
- yake-rust: A port of YAKE!, the statistical keyword extractor from researchers in Portugal, to Rust and WebAssembly, so Node.js could have one that runs at native speed. Node had nothing like it, and Node runs WASM. I translated the Python original to Rust and published to both crates.io and npm.
- Universal iCal for Tidbyt: A calendar app for the Tidbyt retro display that works with any calendar, written in Starlark. Forum users kept asking for one. Every existing app needed OAuth, which many companies block, so I used iCal, the format every calendar already exports. It was merged into Tidbyt’s official community repo and inspired other apps.

## Skills
- Languages: TypeScript, Go, Rust, Python, C#, Ruby, PHP, Starlark
- AI: RAG, vector and semantic search, LLM evaluation, prompt languages, agents
- Systems: Federated GraphQL, microservices, pub/sub, gRPC, serverless, Kubernetes
- Data: Postgres, MongoDB, Redis, Elasticsearch, Spark (Azure Synapse)
- Security: SOC 2, CMMC, authentication and authorization (ABAC, OIDC, Auth0)
- Product: Discovery, design, brand, technical strategy, training and mentoring