Blog

Notes from the lab

Builds, research logs, and what we learned while shipping.

Blueprint Labs · Research

acra: the paper, and where plans start for the last quarter

ACRA is our research architecture for auditable, evidence grounded clinical reasoning. Here is the aim of the paper, why it was never sidelined while we built Leverage, and where the Q4 2026 build plan actually starts.

8 min read

Blueprint Labs · Research

the observation plane: trust boundaries for a chart-reading agent

A chart-reading agent is only as trustworthy as the surfaces that expose it. Here is how we engineered observation planes, auth boundaries, flight recorders, and recoverability around Leverage after Learning.

10 min read

Blueprint Labs · Leverage

leverage playbook memory: shadow mode, evidence, and why we are not learning aggressively yet

Most AI trading tools forget everything or learn too fast from noisy outcomes. Playbook Memory records the full episode graph, runs shadow retrieval first, and only promotes lessons that survive evidence review.

6 min read

Blueprint Labs · Leverage

leverage co-pilot: a trading partner beside the chart

Leverage started as Pulse, an AI scanner that watches charts and surfaces trade ideas. Professional traders need more than signals. Co-Pilot is the second brain we built for that.

5 min read

kiingxo · Research

abstraction all the way up. what two years of building AI systems actually taught me.

Software engineering has always been a story of rising abstraction. AI is the newest layer. Here is what building real systems taught me about where that layer holds and where it breaks.

5 min read

Blueprint Labs · Leverage

leverage, the floor, and why AI tools should feel alive

We built a 3D trading room into Leverage. Here is the thinking behind it, and why making a tool feel alive is not a design indulgence. It is a reliability decision.

3 min read

Blueprint Labs · Leverage

what we cut from leverage in two weeks, and what that taught us about building AI tools for traders

After three weeks of daily use and the first round of beta feedback, we ran a full declutter pass. The lesson was not about design. It was about what happens when you put AI output into a high-pressure environment.

3 min read

Blueprint Labs · Leverage

leverage gets bitcoin. what adding a second market taught us about AI trading context.

Adding BTC support meant confronting something most AI trading tools ignore: different markets need different ground truth, not the same model pointed at a different chart.

2 min read

Blueprint Labs · Leverage

when the model said the stop wasn't hit. it was. what that incident changed about how we build.

The most important architectural decision in Leverage came from one trade going wrong. Here is what we learned about what AI should and should not be deciding in a live trading system.

2 min read

Blueprint Labs · Leverage

leverage: putting AI orchestration beside the gold chart

What it actually looks like to put a full AI orchestration trading system in the hands of someone who trades gold for a living. Not a demo. A tool we use every session.

2 min read

Blueprint Labs · Apex

APEX: open beta ops and how we think about AI at the execution layer

In April we focused on what makes an AI trading system reliable when markets are noisy. The answer is not better AI. It is knowing exactly what the AI should and should not be deciding.

2 min read

Blueprint Labs · Apex

APEX: what quality before scale actually means in an AI trading system

March work on APEX was about raising the bar on candidate call quality before thinking about volume. Here is why that ordering matters.

2 min read

Blueprint Labs · Infrastructure

VPS AI engineer: why alert quality matters more than alert volume, and how we are fixing it

February work on the VPS AI engineer was about reducing noise and anchoring AI suggestions to runbook logic. The goal is a system operators can trust, not one they have to second-guess.

2 min read

Blueprint Labs · Infrastructure

VPS AI engineer: what it actually means to put an AI layer on server monitoring

The first phase of the VPS AI engineer was about giving operators faster context when things go wrong. Not automation. Context, at the speed the situation actually needs.

2 min read

Blueprint Labs · Tooling

turning GitHub activity into something a team can actually read

In December we built a project management layer that summarises what is happening across repos. The problem it was solving is not a GitHub problem. It is an information density problem.

2 min read

Blueprint Labs · Slash

/Slash goes multi-LLM. why single-provider AI is a fragility you do not notice until you need to.

We moved /Slash from a single-provider architecture to multi-LLM routing with OpenRouter support. The change is mostly invisible to users and matters a lot for reliability.

2 min read

Blueprint Labs · Blueprint Labs

introducing blueprint labs

Blueprint Labs engineers AI for hard domains. This is what we are building, why we write in public, and why this blog exists.

2 min read