Built — an autonomous market-synthesis pipeline

Keeping up with my crypto channels was ~11 hours of livestream a week. So I built a machine to do the watching.

It discovers every new episode from Cultivate Crypto and Dollar Cost Crypto, transcribes it, and distills the week into one synthesis — what they think, where they agree, and where they flatly disagree. Refreshed weekly, every claim dated and traceable to the episode it came from.

Auto-published · refreshed weekly · latest source 2026-06-26
~11h
new livestream a week
8 min
to read the week's synthesis
2.5h
in an average episode
62
episodes analyzed so far

What I built

Following crypto on YouTube and Rumble is a part-time job: ~11 hours of livestream a week across just two channels, the signal buried in the noise. So I built a pipeline that does the watching. It discovers new episodes, transcribes them, analyzes each one under a fixed contract, and compounds the result into one living synthesis — a global read plus a per-channel dashboard, refreshed every week.

The synthesis isn't a hot take. It's the durable, attributed, dated state of what the channels are saying — and it keeps itself current.

Consensus and conflict — not a blended average

Two analysts rarely agree, and averaging them into mush loses the most useful signal. The synthesis tags where Cultivate Crypto and Dollar Cost Crypto reach the same read, and shows exactly where they split — so you see the real disagreement (one calls the Fed hawkish while the other reads it as risk-on), not a false consensus.

The hard part of an AI system is keeping it trustworthy

Deterministic
A Python CLI owns every mechanical step — reproducible and testable.
Judgment, on rails
The AI does the thinking, but only inside versioned contracts.
Auditable
Every state change is verified Expected vs Claimed vs Actual.

Start here

This week's synthesis →

The global read and the per-channel dashboards, fresh this week.

How it's built →

The pipeline, the determinism/agent split, and why it's auditable.