Skip to content
View zackproser's full-sized avatar
💭
Studying 📖
💭
Studying 📖

Block or report zackproser

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
zackproser/README.md

Zack Proser

Applied AI @ WorkOS · Independent AI Researcher · Consultant
Local inference economics · Eval methodology · Systems from silicon to prompt

arXiv Blueprints Benchmarks Demos Newsletter Mind on Fire Mind on Fire AI Consultancy


Research

I run controlled experiments on local LLM inference, build eval harnesses with proper statistics, and publish everything as interactive technical drawings.

The Price Floor
The Price Floor
Local inference break-even analysis
M5 Max memory-bandwidth ceilings · Engine tax (llama.cpp vs ds4 = 5×) · ~$34K Blackwell build-or-buy
The Eval Harness
The Eval Harness
Coding agent benchmark (60 runs)
Local vs cloud vs frontier · Hidden-test design · Wilson interval statistics
The Cost Curve
The Cost Curve
DeepSeek V4 Flash economics
MoE routing (284B/13B active) · Quantization ladder · Cache-hit dominance (96.6%)
The Inference Engine
The Inference Engine
Serving architecture deep-dive
Prefill/decode · KV cache & PagedAttention · SARATHI · DistServe
The Benchmark
The Benchmark
Measurement methodology
MMLU · SWE-bench · Chatbot Arena · Goodhart's law in eval
Practical Local Inference
Practical Local Inference
M5 Max + DeepSeek V4 Flash at 30-40 tok/s
Workload routing matrix · $17/mo via Gateway

The Blueprint Series

17 technical drawings covering the full AI stack — each with SVG schematics, equations, academic citations, and interactive demos:

The Transformer
The Transformer
The Tokenizer
The Tokenizer
The Embedding Space
The Embedding Space
The RAG Pipeline
The RAG Pipeline
The Diffusion Model
The Diffusion Model
Choosing an LLM
Choosing an LLM
The Agent Fleet
The Agent Fleet
The Attention Head
The Attention Head
The Autonomy Boundary
The Autonomy Boundary
The Guard
The Guard
Designing AI Evaluations
Designing AI Evaluations
The Workshop
The Workshop

→ View all 17 Blueprints


Interactive Demos

Live tools you can use right now — built into my portfolio site:

Local vs Cloud Benchmark
Local vs Cloud Benchmark
60-run evidence explorer
Tokenization Lab
Tokenization Lab
BPE pricing calculator
RAG Visualized
RAG Visualized
Pipeline step-through
Embeddings Explorer
Embeddings Explorer
3D vector space
Voice AI Pipeline
Voice AI Pipeline
Meeting intelligence


Firecrawl, vertical voice AI,
knowledge bases & more

Publications

  • Identity Management for Agentic AI (Oct 2025) — Co-authored white paper on authentication, authorization, and security for AI agents. With Tobin South, Alex Pentland (MIT), and others.
  • RAG Evaluation chapter — Official contribution to Pinecone's Vector Databases in Production book series.
  • 50+ technical articles at WorkOS, Pinecone, Cloudflare, Gruntwork — AI agents, RAG pipelines, fine-grained auth, vector DBs, IaC.
  • 180+ essays on zackproser.com + external publications on The New Stack, Prefect.io, Pinecone Learning Center.

→ Full publications list


Open Source

Project Description Stars
coding-agent-eval-harness Open-source benchmark harness for coding agent comparison NEW
portfolio zackproser.com — Next.js 15, 17 Blueprints, 9 demos, 180+ essays
handwave watchOS app controlling Claude Code from your wrist
git-xargs Run commands across many GitHub repos simultaneously ⭐ 1,086
cloud-nuke AWS resource cleanup — nuke entire accounts safely ⭐ 3,121
cf-terraforming Generate Terraform from existing Cloudflare config ⭐ 1,324

Speaking

DevSecCon 2025 Keynote
DevSecCon 2025
Keynote: Walking and Talking in the Woods with AI
Voice-first development · Agent autonomy
Claude Cowork Workshop
WorkOS × Anthropic
Claude Cowork GTM Workshop
800 registrations · Hands-on build with Lydia from Anthropic
Skills at Scale Workshop
AI Engineering London
Skills at Scale Workshop (80 min)
Claude Code skills across workflows & teams
Untethered Productivity Talk
AI Engineering London
Untethered Productivity
Staying healthy, creative & shipping in the AI era
Applied AI Showcase
WorkOS Applied AI Showcase
Three Learnings from Shipping
Interface beats stack · Complete the loop
a16z Pinecone Talk
a16z San Francisco
From Jupyter Notebooks to Production
Pinecone AWS Reference Architecture · ~125 attendees

→ All speaking engagements


Career

Role Company Years What I did
Applied AI WorkOS 2025–present AI workflows, internal tooling, developer education. Created Claude Cowork workshop with Anthropic.
Staff DevRel Pinecone 2023–2025 RAG pipelines, AWS reference architectures, dev education during the vector DB explosion. Spoke at a16z SF.
Senior Software Engineer Cloudflare 2020–2023 Joined at ~100 engineers globally. Built cf-terraforming (1.3K⭐). APIs, edge infra, dev tools at scale.
Infrastructure Engineer Gruntwork 2018–2020 Built git-xargs (1K⭐), contributed to cloud-nuke (3.1K⭐). Multi-account AWS, Terraform, production IaC.

Reach

12.4M

Search impressions / yr
Google Search Console

52.5K

Search clicks / yr
avg. position 7.9

137K

Visitors / yr
+21% YoY

179K

Page views / yr
+18% YoY

6,000+

Newsletter
subscribers

Regularly advising hedge funds and PE firms on AI infrastructure, developer tools, and cloud platforms via AlphaSights, Guidepoint, Tegus, GLG, Coleman.


zackproser.com · Twitter/X · LinkedIn · YouTube · Mind on Fire mindonfire.net

Pinned Loading

  1. automations automations Public

    Shell scripts that leverage the latest AI models and github.com/charmbracelet tooling to make your developer workflow buttery smooth and way more fun

    Shell 76 4

  2. gruntwork-io/git-xargs gruntwork-io/git-xargs Public

    git-xargs is a command-line tool (CLI) for making updates across multiple Github repositories with a single command.

    Go 1.1k 73

  3. cloudflare/cf-terraforming cloudflare/cf-terraforming Public

    A command line utility to facilitate terraforming your existing Cloudflare resources.

    Go 1.4k 169

  4. CanyonRunner CanyonRunner Public

    A complete HTML5 game using the Phaser framework. Playable, buildable, and published as a tutorial for other game developers.

    JavaScript 134 24

  5. goodneighbor goodneighbor Public

    Twitter AI Platform

    JavaScript 95 17