The signal in
the noise.

Feza builds autonomous trading systems that find, validate, and execute alpha across global markets — powered by ML infrastructure we built ourselves, for exactly this.

4 Validated Models
>1.0 Sharpe (all models)
245 tok/s Inference
950+ Tests Passing

Markets reward intelligence,
not intuition.

Every competitive advantage in trading eventually becomes infrastructure. We're building the infrastructure from day one.

AI is the edge

Financial markets are information-processing machines. The participant with the best signal extraction wins. Our systems process market data continuously, finding patterns human traders cannot see and acting on them without hesitation or emotion.

Vertical integration

We built our own ML framework. We control the full stack from tensor operations to order execution — no dependency on Python runtimes, no vendor lock-in, no performance ceilings we can't break through ourselves.

Compounding advantages

Technology improves. Track records lengthen. Data accumulates. Relationships deepen. Every quarter we operate, the business becomes harder to replicate. We're not chasing a single edge — we're building a machine that finds the next one.

Two systems. One objective.

One system trades. The other thinks. Together, they form a vertically integrated trading intelligence that improves itself continuously.

Trading Engine

Proprietary · Autonomous

Our trading system ingests market data, runs model inference, enforces risk limits, and executes orders — autonomously. The full signal-to-order loop runs without human intervention.

Every model is validated before it trades. Automated retraining keeps strategies current. Real-time risk controls protect capital around the clock.

Go Distributed Real-time Autonomous

Zerfoo

Open Source · ML Framework

A production-grade ML inference and training framework written entirely in Go with zero CGo dependency. Runs 6 model architectures at 245 tok/s — 20% faster than Ollama. Every trading signal we generate flows through Zerfoo inference.

Open source under Apache 2.0. 25+ custom CUDA kernels. OpenAI-compatible API. The Go ecosystem's answer to PyTorch.

Pure Go CUDA GGUF ARM NEON Purego Apache 2.0

One language. One codebase. Zero compromise.

Most quant firms stitch together Python for research, C++ for execution, Java for infrastructure, and pray the interfaces don't break. We write everything in Go — from CUDA kernel bindings to REST APIs to model training.

Single binary deployment. No runtime dependencies. No GIL. No framework lock-in. When we find a bottleneck, we fix it in the same language the rest of the system speaks.

// The loop
data       → ingest & normalize
features    → engineer & select
inference  → predict direction
risk       → validate & size
execute    → order → fill
learn      → retrain & promote
// Every step: Go. Every µs: counted.

Proof, not promises.

Every model passes through a rigorous validation pipeline before it touches real capital. No lookahead bias. No curve fitting. No exceptions.

4
Validated Models
>1.0
Sharpe Ratio (all models)
<15%
Max Drawdown Target
950+
Tests Passing

A decade of compounding.

We think in decades. Each phase builds the foundation for the next — expanding into new markets and compounding both capital and capability.

Current
2026

First Capital

Live trading with real capital. Automated model retraining. Real-time risk controls. Proving P&L in production conditions.

Target: Sharpe > 1.0, positive cumulative P&L
2027–2028

Expansion

New asset classes. Multi-strategy portfolio management. 12-month audited track record. Institutional-grade reporting.

Target: Multiple uncorrelated strategies
2029–2031

Scale

Fund launch. Global markets. The system expands into new territories and begins discovering strategies autonomously.

Target: Multi-asset class, multi-geography
2032–2036

Apex

Full-spectrum market coverage. The world's most valuable trading intelligence — built by machines, for machines.

Target: Global, autonomous, compounding

Our ML engine is open source. On purpose.

Zerfoo is the ML inference framework that powers every trading signal we produce. We open-sourced it because the best infrastructure gets built in the open — by the community, for the community.

A stronger foundation benefits everyone. What we build on top of it is ours.

Explore Zerfoo →
// One line. That's it.
model, _ := zerfoo.Load("google/gemma-3-4b")
 
// 6 architectures. 245 tok/s.
// 25+ CUDA kernels. Zero CGo.
// 20% faster than Ollama.
 
resp := model.Chat(ctx, messages)
// OpenAI-compatible. Drop-in.
"I spent eight years watching human emotion destroy perfectly good trading strategies — including my own. So I decided to remove the human from the loop entirely."

David built distributed systems at scale at Zendesk, then taught himself machine learning from first principles. Feza is not a side project or an experiment — it's the convergence of deep infrastructure expertise and an obsessive conviction that markets belong to machines.

One founder. A fleet of AI agents. Building the most automated trading company on earth.

The signal is clear.

We're building in public, shipping in private, and compounding from day one.