live
↗
This server (AI-native resume)
The MCP server you are querying right now.
0 runtime dependencies · 11 tools · one of them shows you dolphins
an AI-native resume
Eric Backman
Data Engineer & Applied AI, Agentic Systems · Toronto, ON, Canada
A resume can claim anything. This one is built as evidence: it is served by an MCP server Eric built, its narrative was written by the AI he works with every day, and every number in it was pulled from the workspace it describes.
27
repos with real history (of 87 on disk)
1,035
commits in 2026
365
trust-scored agent sessions
39
hostnames on the estate
4
MCP servers built
25
operational playbooks
Numbers verified against the workspace on 2026-08-10 — counted from data files, not READMEs.
#Query this resume
This terminal talks to the real MCP endpoint on this domain. Click a tool. What your AI assistant would see is exactly what you'll see.
#Watch the output
Live from the channel the Scuba Sessions studio runs. Eric shot every frame; Claude reviewed the footage, cut and color-corrected each video, uploaded it, and scheduled its release. Tap to watch on YouTube.
A reef shark on patrol
The pod just kept coming
Gliding with a manta ray
A wall of barracuda
Swimming with dolphins in 4K
More at @scubasessions — 90 videos public, all cut and scheduled by the pipeline.
#177 days
Outside a portfolio page from 2017 and one archived stats folder, every repository in the workspace has a first commit on or after 2026-02-23. First sports database to this server: 155 days. Seven tracks, each one a question that kept getting answered. Your AI can read the same story via the get_timeline tool.
#The system map
The shipped projects, with their moving parts. Click any card to ask the server about it. A live badge means you can visit it right now; scheduled means it runs unattended on a timer.
live
↗
Scuba Sessions studio (YouTube + Instagram)
Tools that let Claude turn a pile of raw dive footage into a running YouTube channel and an Instagram feed.
90 videos public · 93,220 channel views · 500 source clips catalogued · 217-entry publish queue
footage intakeeditorupload & scheduleanalytics monitorlong-form crewsocial manager
live
Agent reliability dashboard
Scores every logged Claude Code session on how far it ran unattended.
365 logged sessions across 47 projects · 33,137 tool calls and 7,479 files changed measured · average trust 81/100, correction rate 0.21
live
↗
Agentic paper trader
Claude runs a trading session every weekday morning and publishes its reasoning.
90-day public decision log
trading enginereport worker
complete
Claim-locked video essays
Data-driven video essays where a fabricated number is structurally impossible.
every on-screen number traces to a data row · majors history to 1960
claim engineadversarial reviewself-hosted voicechart renderer
complete
Local-model agent sandbox
A boundary that makes it safe to let an agent run unattended on hardware you already own.
639x faster prefill with prompt caching · 8.6 tokens/sec on CPU with no usable GPU · 0 credentials inside the sandbox
live
Agentic-commerce conformance prober
Checks whether a store's machine-readable storefront is actually valid, before a shopping agent trips over it.
228 tests · 19 storefronts surveyed, 14 serving a valid profile · all 14 still exposed to a sunsetting endpoint
profile validatortransport probemigration diffagent-profile monitor
scheduled
Review-gated clip pipeline
Turns a folder of footage into vertical clips, where the thing you approved is literally the thing that ships.
203 tests · 0 runtime dependencies
complete
Pre-registered commerce experiment
A $500 business experiment run like a study, with the money ledger in public.
$500 hard cap, $12.20 spent so far · 4-stage KPI tree, every gate pre-registered
live
↗
AskViz
Ask a plain-English question about your data, get a chart on your phone.
19 passing tests · ~20 KB gzipped bundle
complete
Sports data platform
Local multi-sport SQLite databases built to answer any sports question in plain English.
7M-row golf holes table · NBA box scores 1946-present · NFL 1999-present · NHL 1997-present
NBANFLNHLGolfsports-data MCP serversports-cronspodcastsData Lab
scheduled
Job-hunt operating system
The job search itself, run as an agentic system.
28 job sources swept daily · 4 purpose-built subagents
role-scouttracker-reconcilerresume-tailorvoice-ats-reviewerdaily-jobs-watchliveness watchdog
live
Self-writing engineering newsletter
A weekly newsletter that reads his AI's work logs and writes itself.
7 published issues · ~100 MB of transcripts condensed to ~2 KB per issue
live
FRM Level 1 study system
A zero-inference personal tutor for the FRM Part I exam.
501 questions across 12 banks · 60 textbook-chapter note sets
live
↗
Interactive dive map
A world map of every scuba dive Eric has logged.
147 dives across 20 trips (counted from the data file)
live
↗
The $15 Gauntlet
Build a $15 team, go 16-0. A game family with a tested difficulty contract.
500-grid simulated difficulty contract · 3 sport variants from 1 template
NBAMLBNHL
live
↗
Discord bot fleet
A canonical Worker bot template cloned into 5 live prediction and clan bots.
5 bots from 1 canonical template
picks-workersumo-picksf1-picksosrs-clan-botwow-clan-bot
complete
Life Tracker, MCP server + personal data platform
Personal data platform with an MCP server exposing read-only query tools to Claude.
5 read-only MCP tools
complete
↗
Side Bet
A multiplayer party game, built two-person with a protected-main PR workflow.
28 unit tests on the core betting logic
#Ask your AI about Eric
The honest screen: the server lists what he hasn't done alongside what he has. Two ways in; the first needs zero setup.
No setup, any AI. Paste this into Claude, ChatGPT, or anything that reads a URL:
Read https://ericbackman.com/resume.md and https://ericbackman.com/llms.txt,
then tell me whether Eric fits this role: [paste the job description]MCP connector, if you want the real thing. Endpoint: https://ai.ericbackman.com/mcp
Claude: Settings → Connectors → Add custom connector. Claude Code:
claude mcp add --transport http eric-backman https://ai.ericbackman.com/mcpAny MCP client:
{ "mcpServers": { "eric-backman": { "type": "http", "url": "https://ai.ericbackman.com/mcp" } } }Your assistant may ask you to approve the first tool call — that's your AI being properly cautious with an unfamiliar server, not something this server can waive. Every tool is read-only and says so in its MCP annotations. One approved call to about returns the complete brief; approving it "always" makes the rest seamless.
Human? The prose version is at /resume.md — a narrative resume written by Claude, in its own voice, from inside Eric's workspace.
Crawler? /llms.txt.
This page, the server, and the pipeline that verified its numbers are all part of the work sample: Hand-rolled JSON-RPC over Streamable HTTP on a Cloudflare Worker. No runtime dependencies. Stateless. The content lives in one typed data module.