PRACTICE / 02 — AI LABS

Research that turns into product.

Labs is our internal R&D arm. We use it to keep practical knowledge ahead of client work, and the work occasionally ships as standalone systems we run ourselves.

What we ship into
ResearchSpikeEvalShipOperateOpen-sourceResearchSpikeEvalShipOperateOpen-source
Lab products

Things we built that we still run.

Not demos. Running systems on our own infrastructure, under the same standard we sell to clients.

LAB · PRODUCT

ProxiAlpha

Agent-driven trading research platform. Walk-forward strategy research, paper-trading, and a live cockpit. End-to-end instrumented and continuously evaluated.

STRATEGIES
12+
UPTIME
99.9%
EVALS
Daily
See the product
LAB · PRODUCT

ProxiTrades

The execution and routing layer ProxiAlpha sits on top of. TradingView webhook receiver, strategy registry, paper and live brokers, full audit log.

BROKERS
3
LATENCY
~80ms
AUDIT
Full
See the product
LAB · LIVE DASHBOARD

ProxiAlpha Terminal

Public-readable view of the trading research platform. Live signals, walk-forward folds, portfolio metrics, and a candlestick view of every active strategy. Useful both as a working dashboard and as a reference for how we build operator-grade UI.

Open the terminal
terminal · live
$ proxi status
● strategy_v7 live sharpe 1.42
● mean_reversion paper sharpe 0.91
◐ momentum_x eval pending
· last update 23s ago
$ proxi tail signals --strategy strategy_v7
10:42:18 BUY AAPL conf=0.83 size=0.7%
10:43:51 HOLD MSFT conf=0.61
10:45:09 SELL NVDA conf=0.79 size=1.2%
Research areas

Where we spend our time.

  • AREA / 01

    Agent systems

    Tool design, planning, multi-step reasoning, long-horizon tasks. The hard parts of agent reliability.

  • AREA / 02

    Evaluation tooling

    Eval harnesses, regression tests, drift detection. The instrumentation that keeps systems honest in production.

  • AREA / 03

    Quantitative trading

    Walk-forward research, regime detection, ensemble strategies. ProxiAlpha is our test bed.

  • AREA / 04

    Retrieval & RAG

    Chunking, embeddings, hybrid search, and the failure modes that show up at scale.

  • AREA / 05

    Red-teaming

    Adversarial probes, prompt injection, jailbreak defenses, and a working protocol that ops teams can run.

  • AREA / 06

    Operator-grade UI

    UIs for the people running AI in production. Dense, fast, opinionated. The terminal is the running example.

Why we run a lab

Practitioners teach better.

A consulting team that doesn't build will eventually start advising from secondhand knowledge. Once that happens, you start hearing things that sound right and don't survive contact with production.

Labs is how we keep that from happening. We run our own systems, on our own infrastructure, under the same operating standard we sell to clients. The lessons from those systems show up in client engagements directly.

The training practice draws from the same well — the curriculum is built on what we've actually shipped, not on what's currently trending.

Lab loop
ResearchSpikeEvalShipOperateOpenResearchSpikeEvalShipOperateOpen

Want to collaborate?

Labs occasionally takes on research collaborations and partnerships when the work is a fit. Tell us what you have in mind.