You're viewing v1 — the original design (early 2026). Switch back to v2 top-right for what Emi is today.
Products · v3

What the factory
has shipped.

Four live products and one beta preview, five different domains — motorsport telemetry, soccer tactics, the fleet's own cockpit, real-world social connection, and open-source open intelligence — all shipped by the same factory. The lessons compound: quality gates Apex earned the hard way are blocking from the first commit on every product after.

New in v3: azimuth published its whole knowledge graph as an explorable linked-data surface, and Mission Control grew three governance panels — the fleet's self-change record, readable in one place.

01
F1 telemetry · live
Live · Product 1

Apex

F1 telemetry, in your browser. Every session, every driver, every lap — speed, throttle, brake, gear, RPM, lap deltas, championship progression.

Race-engineer AI on every chart. Claude writes a breakdown per diagram: what changed in the stint, why the tyre call was right or wrong, where the next gain is.

Tech stack

Frontend
Next.js 16 · React · Recharts · Vercel
Backend
FastAPI · Hetzner + Coolify
Data
FastF1 · OpenF1 · multi-season fixtures
AI layer
Claude — race-narrative analysis, model pinned
~12 days
Start → first users
May '26
Live since
255
Git commits
33,742
Lines of code
~$60
Estimated build cost
apex.emi-factory.dev
Apex — F1 telemetry with race-engineer AI analysis
● LIVE
02
soccer KG · live
Live · Product 2

Corner

Live knowledge graph of what a team actually did on the pitch. Every pass, duel, press and recovery becomes a graph edge. The picture of how the team is really connected builds itself in real time.

Predict, then measure. Pre-match, a coach scores team strength from the formation. Post-match, the graph replays what actually happened from real event data. The gap between the two is the coaching insight.

Tech stack

Frontend
React + Vite · Cytoscape.js · Tailwind
Backend
Express 5 (Node 24) · Hetzner + Coolify
Data
StatsBomb event data · World Cup 2026 squads
AI layer
Claude — coach insights at the largest delta
Live
Status
41,639
Lines of code
444
Git commits
~$25
Estimated build cost
03
fleet cockpit · live · internal
Live · Internal · Product 3

Mission Control

A live cockpit for the agent fleet. Task queue, weekly goals by track, recent activity — everything Emi is doing right now, on one screen. Not a log file, a real dashboard built for the operator.

New in v3 — the governance layer, readable. A panel of how well the fleet predicts Michael's own decisions, the kill-switch registry (one master switch stops everything), and metric tiles that show UNKNOWN — not a comforting zero — when a source goes stale.
Private · access-locked — Michael only
Live
Status
Real-time
Fleet reads
12
Tracks monitored
~$0
Hosting cost
mission control · private
Mission Control — the fleet's live cockpit (sandbox data shown for privacy)

Sandbox data shown for privacy — the real cockpit is access-locked.

04
open intelligence · live · open source
Live · Open source · Product 4

azimuth

An open-intelligence vault that runs itself. A daily GitHub-Actions job pulls open WorldMonitor data into dated source notes; an AI curator synthesises them into evolving weekly briefs under a machine-enforced editorial contract; a cross-channel knowledge graph answers questions across energy, geophysics, climate and more. Emi's own vault doctrine, demonstrated in public.

Governance you can check — and now a graph you can browse. Every claim in a brief links to the exact source note it rests on, the editorial line is enforced by blocking CI lints, and new in v3: the whole vault — concepts, sources, briefs — is published as an explorable linked-data (RDF) graph, regenerated with every ingest day. The whole engine is a public, open-source repo.

Tech stack

Engine
Pure standard-library Python · GitHub Actions
Site
Deterministic static build
Data
WorldMonitor open-intelligence API · licensed sources only
AI layer
Claude curator — weekly synthesis under a CI-lint contract
Live
Status
Open
Source — public repo
Daily
Automated data ingest
Weekly
AI-curated briefs
RDF
Explorable linked-data graph
05
real-world social · beta preview
Beta preview · Product 5

Echo

Mutual-spotting, real-world connection. Describe someone you saw in public. If they describe you back, both sides can connect — recognition has to be reciprocal, so there is no one-sided messaging.

Two descriptions, one match. Every description becomes a vector embedding; a match only fires when two people's accounts of each other line up semantically. The reciprocity is the privacy model — you only surface to someone who already saw you too.
Not live — in beta preview

Tech stack

Frontend
Next.js 16 · React 19 · TypeScript · Vercel
Backend
FastAPI · Hetzner + Coolify
Data
Postgres + pgvector — semantic match store
AI layer
Text embeddings — reciprocal similarity matching
Beta
Status
15,929
Lines of code
74
Git commits
Demo
Auth posture
echo · beta preview · not public yet
Echo — mutual-spotting real-world connection app (beta preview)
● BETA
← Home
04 · Products

What the factory
has shipped.

Four live products and one beta preview, five different domains — motorsport telemetry, soccer tactics, the fleet's own cockpit, real-world social connection, and open-source open intelligence — all shipped by the same factory. That's the point: the lessons compound. Every project teaches the next one. Quality gates Apex earned the hard way are blocking from the first commit on every product after.

New since v1: Corner went live, Mission Control became the fleet's own cockpit, azimuth went live as the first fully open-source product, and Echo (a mutual-spotting connection app) shipped in beta preview — four live now, up from one, plus one in beta.

01
F1 telemetry · live
Live · Product 1

Apex

F1 telemetry, in your browser. Every session, every driver, every lap — speed, throttle, brake, gear, RPM, lap deltas, championship progression.

Race-engineer AI on every chart. Claude writes a breakdown per diagram: what changed in the stint, why the tyre call was right or wrong, where the next gain is.

Tech stack

Frontend
Next.js 16 · React · Recharts · Vercel
Backend
FastAPI · Hetzner + Coolify
Data
FastF1 · OpenF1 · multi-season fixtures
AI layer
Claude Sonnet — race-narrative analysis
~12 days
Start → first users
89 days
Live and running
255
Git commits
33,742
Lines of code
~$60
Estimated build cost
apex.emi-factory.dev
Apex — F1 telemetry with race-engineer AI analysis
● LIVE
02
soccer KG · live
Live · Product 2 now live

Corner

Live knowledge graph of what a team actually did on the pitch. Every pass, duel, press and recovery becomes a graph edge. The picture of how the team is really connected builds itself in real time.

Predict, then measure. Pre-match, a coach scores team strength from the formation. Post-match, the graph replays what actually happened from real event data. The gap between the two is the coaching insight.

Tech stack

Frontend
React + Vite · Cytoscape.js · Tailwind
Backend
Express 5 (Node 24) · Hetzner + Coolify
Data
StatsBomb event data · World Cup 2026 squads
AI layer
Claude Haiku — coach insights at the largest delta
Live
Status
41,639
Lines of code
444
Git commits
~$25
Estimated build cost
03
fleet cockpit · live · internal
Live · Internal · Product 3 new

Mission Control

A live cockpit for the agent fleet. Task queue, weekly goals by track, recent activity — everything Emi is doing right now, on one screen. Not a log file, a real dashboard built for the operator.

The fleet watches itself. Mission Control reads the fleet's own live data — the queue, the progress, the dispatches. Every refresh is a current snapshot.
Private · access-locked — Michael only
Live
Status
Real-time
Fleet reads
12
Tracks monitored
~$0
Hosting cost
mission control · private
Mission Control — the fleet's live cockpit (sandbox data shown for privacy)

Sandbox data shown for privacy — the real cockpit is access-locked.

04
real-world social · beta preview
Beta preview · Product 4 not live yet

Echo

Mutual-spotting, real-world connection. Describe someone you saw in public. If they describe you back, both sides can connect — recognition has to be reciprocal, so there is no one-sided messaging.

Two descriptions, one match. Every description becomes a vector embedding; a match only fires when two people's accounts of each other line up semantically. The reciprocity is the privacy model — you only surface to someone who already saw you too.
Not live — in beta preview

Tech stack

Frontend
Next.js 16 · React 19 · TypeScript · Vercel
Backend
FastAPI · Hetzner + Coolify
Data
Postgres + pgvector — semantic match store
AI layer
Text embeddings — reciprocal similarity matching
Beta
Status
15,929
Lines of code
74
Git commits
Demo
Auth posture
echo · beta preview · not public yet
Echo — mutual-spotting real-world connection app (beta preview)
● BETA
05
open intelligence · live · open source
Live · Open source · Product 5 new

azimuth

An open-intelligence vault that runs itself. A daily GitHub-Actions job pulls open WorldMonitor data into dated source notes; an AI curator synthesises them into evolving weekly briefs under a machine-enforced editorial contract; a cross-channel knowledge graph answers questions across energy, geophysics, climate and more. Emi's own vault doctrine, demonstrated in public — architecture without the private content.

Governance you can check. Every claim in a brief links to the exact source note it rests on, the editorial line is enforced by blocking CI lints — not guidelines — and the knowledge graph ranks cross-channel connections by source evidence. Answers a static data feed cannot give. The whole engine is a public, open-source repo.

Tech stack

Engine
Pure standard-library Python · GitHub Actions
Site
Deterministic static build · Cloudflare Pages
Data
WorldMonitor open-intelligence API · licensed sources only
AI layer
Claude curator — weekly synthesis under a CI-lint contract
Live
Status
43 days
Running itself
756
L1 source notes
15
Briefs maintained
~$3.66
LLM spend, all-time
← Home
04 · Products

What the factory
has shipped.

Each product is a different domain — motorsport telemetry, soccer tactics, more on the way — but the same factory ships them. That's the point: the lessons compound. Every project teaches the next one. Quality gates Apex earned the hard way are blocking from the first commit on every product after.

The factory is a blueprint, not a one-off. The same pipeline that ships an F1 telemetry analyser ships a soccer knowledge graph ships whatever the third product turns out to be. Speed and quality grow with each iteration because the foundation grows with each iteration — skills, scaffolds, deploy plumbing, review patterns, all transferable.

01
F1 telemetry · live
Live · Product 1

Apex

F1 telemetry, in your browser. Every session, every driver, every lap — speed, throttle, brake, gear, RPM, lap deltas, championship progression.

Race-engineer AI on every chart. Claude writes a breakdown per diagram: what changed in the stint, why the tyre call was right or wrong, where the next gain is. Not available anywhere else.

Tech stack

Frontend
Next.js 16 · React · Recharts · Vercel
Backend
FastAPI · Hetzner + Coolify
Data
FastF1 · OpenF1 · multi-season fixtures
AI layer
Claude Sonnet — race-narrative analysis
~12 days
Start → first users
25 days
Start → today
161
Git commits
21,315
Lines of code
~$60
Estimated cost (Claude Plan tokens)
apex.emi-factory.dev
Apex — F1 telemetry with race-engineer AI analysis
● LIVE
02
soccer KG · shipping in days
Shipping in days · Product 2

Corner

Live knowledge graph of what your team actually did on the pitch. Every pass, duel, press and recovery becomes a graph edge. The picture of how the team is really connected builds itself in real time.

Predict, then measure. Pre-match: Tactical Planner scores team strength from formation data. Post-match: Knowledge Graph Replay shows the real score from StatsBomb event data. The gap is the coach's diagnostic.

Closed-build phase. Target: first beta coaches in the next 10 days.

Tech stack

Frontend
React + Vite · Cytoscape.js · Tailwind
Backend
Express 5 (Node 24) · Hetzner + Coolify
Data
StatsBomb open event data — World Cup 2022
AI layer
Claude Haiku — coach insights at the largest delta
10 days
Target start → first users
2 days
Factory takeover age
35
Git commits so far
20,437
Lines of code
~$15
Estimated cost (Claude Plan, projected)
corner · closed beta
Corner — soccer knowledge graph pitch replay
03
third product · in design
In design · Product 3

Already on the pipeline

A consumer app the factory will ship next. Different sport, different audience, same standard. Spec written, brand identity sketched, build pending — slotted into the factory roadmap behind Corner.

Details under wraps until first preview. The point is: the pipeline is full, not empty.

Spec'd
Status
~6 weeks
Build window
2026 Q3
Target launch
PRODUCT 3
In design
Details under wraps