← Flagship systems

Independent side project · Solo Builder

Research discipline before performance theater.

Curren is where I treat quantitative research, live-system state, and public product surfaces as different authorities. The goal is a system that can preserve what was known, what was tested, what is running, and what may safely be shown — without publishing the alpha itself.

StartedJun 2026
RoleSolo Builder
Claim boundaryNo guaranteed or validated profitable alpha claimed

The engineering problem

A backtest, a runtime, and a public dashboard should not share authority by accident.

Quant systems fail quietly when research can look into the future, when experiment state mutates after selection, when live logic drifts from historical logic, or when a public surface starts exposing private trading state.

I built Curren around explicit evidence and state boundaries so those failure classes are visible before they become performance claims.

Public system cut

Evidence in. Authority stays separated.

This diagram is intentionally capability-level. It omits source identities, feature definitions, research parameters, private repository topology, exchange/account controls, credentials, and production host details.

01

Point-in-time evidence

Market evidence is treated as time-bounded source material rather than a mutable spreadsheet of hindsight.

02

Research & evaluation

Hypotheses, features, validation, and selection are kept reproducible and separated from later runtime state.

03

Guarded runtime

Signal lifecycle, risk, execution boundaries, reconciliation, and restart behavior are explicit system responsibilities.

04

Public verification

Only a sanitized, read-only projection crosses into the public product surface. Private research and trading authority stay upstream.

What this demonstrates

Systems thinking across research and runtime.

Causality.

Point-in-time evidence and versioned computation keep hindsight from quietly entering research inputs.

Evaluation discipline.

Selection and validation are treated as governed processes rather than a single attractive backtest.

Research/runtime parity.

Historical, replay, and streaming paths are designed around shared semantics instead of independent copies of strategy logic.

Operational state.

Lifecycle, risk, reconciliation, failure, and restart behavior are first-class concerns around any execution path.

What is public

A deliberately boring read-only edge.

The public `sangtrx/curren` repository exposes a sanitized developer surface: read-only API/client/CLI/MCP contracts and verification-oriented projections. It is intentionally not the strategy generator or private trading control plane.

View public repository ↗

Deliberately withheld

The alpha is not the portfolio artifact.

Private source identities, feature definitions, signal parameters, model internals, execution credentials, account state, trading databases, operational controls, hostnames, machine names, and unreleased performance evidence are excluded. A technical interviewer should be able to judge the engineering without receiving a blueprint for the strategy.

Statistical evidence, execution readiness, and live profitability are different claims.

Curren is designed to keep them different. The public case study proves the discipline and system boundaries, not an investment return.

Explore more

See the broader AI systems work.