Skip to content
Aditya Dalmia
All projects

Helmsman

A platform for research, experiments, and the work between them.

Helmsman brings together research engines, shared data infrastructure, an operator workspace, and Structure Lab. Built independently, with a record of what we tried and what we learned.

My work
Co-building the platform, architecture and engineering
Current state
Private platform
Source access
Private
Request a Helmsman walkthroughExplore the evidence
HelmsmanSystem map
Shared evidence. Independent experiments.

The platform

Helmsman is our independent trading research platform. It brings data capture, research, evaluation, monitoring, and an operator workspace into one broader project. The engine that interprets evidence is one component inside that larger system.

Different jobs, clear boundaries

The engine interprets captured evidence. The operator workspace gives research ideas, monitors, decisions, and audits a shared home. Structure Lab turns visual pattern definitions into versioned, testable configuration. Shared infrastructure connects them without erasing those responsibilities.

How it evolved

The work expanded from one engine toward named variants over a common data layer. It includes stored-output replay, forward observation, and a deliberate distinction between changing calculations and changing a model’s interpretation contract.

The parts we set aside

The project history includes Polymarket experiments, rejected graph-based assumptions, a decision against deep historical model re-scoring, and a scheduler migration. Those decisions are part of the build, with their own articles below.

A private walkthrough

Helmsman is privately operated. Its workspace has a deployed, protected web interface; the platform continues to evolve. Contact me for a walkthrough and to discuss access.

What each part is responsible for.

The stack matters at the boundaries between components.

Python
Research engines interpret captured inputs and replay stored outputs; deterministic code owns calculations.
TypeScript · Next.js · Hono
The Helmsman workspace separates the operator interface from its API and state transitions.
PostgreSQL
Stores research records and workspace state, with transactions around selected multi-object changes.
Kubernetes · CronJobs
Runs the engine’s source-controlled schedules. Desired configuration remains distinct from a verified run.

Inspect the work.

Engineering accounts explain the implementation, tradeoffs and limits. Private source is not available through this site.

Private research platform

Experimental; current data provenance and complete live research workflows remain unverified.

Explore
Guided walkthrough on request
Source
Private
Content last reviewed

Several systems. One platform.

The responsibilities behind Helmsman.

Shared data layer

Captured inputs, explicit timestamps, revisions, and data-quality states.

Research engines

Versioned interpretations and deterministic rules that can be compared over shared evidence.

Operator workspace

The workspace for research ideas, monitoring, decisions, and audit history.

Structure Lab

A place to turn visual pattern definitions into explicit, testable configurations.

Evaluation and replay

Recorded model outputs, reproducible calculations, and a distinction between historical replay and forward observation.

Operations

Scheduled jobs, health checks, controlled releases, and a durable trail of changes.

See Helmsman from the inside.

Helmsman is a privately operated trading research platform. Get in touch for a walkthrough and to discuss access.

Request a Helmsman walkthrough

The build, in writing.

10 notes on the architecture, experiments, and tradeoffs.

Read the series
  1. 01Building Helmsman as a platformHow one research engine grew into a shared data layer, an operator workspace, and a place to test competing ideas.7 min
  2. 02A data architecture that remembers what it knewOne synthetic corrected value traced through the real observation writer and schema: revisions as new rows, an as-known-at query, frozen run inputs, and freshness judged against schedule and consumer.8 min
  3. 03Where the model stops and the code takes overThe contract between model output and deterministic code in Helmsman, traced with synthetic replies: what a model may emit, what the validator rejects, and what happens when the handoff fails.9 min
  4. 04Score once, experiment many timesA synthetic identity record traced through the real score store: which change needs new model work, which only reruns deterministic math, and why replaying stored scores does not recover the past.8 min
  5. 05Helmsman: from research candidate to monitorStable identity, evidence gates, and visible deployment states in the Helmsman operator workspace.8 min
  6. 06Turning a chart argument into a testable definitionOne vague chart claim turned into three versioned definitions and replayed on synthetic data through Structure Lab's real detector and registry. The replay shows what a new version does to earlier annotations and evidence.8 min
  7. 07What we kept from the experiments we set asidePolymarket research, unreliable causal graphs, and the value of preserving a rejection with its evidence.3 min
  8. 08Building a real-time news parsing pipelineOne synthetic headline traced through a partial observation, an upgrade, duplicate deliveries, an edit and a retraction, using the ingestion writer's real storage contract.6 min
  9. 09Moving the scheduler without running the work twiceA cutover trace from Helmsman's move to Kubernetes CronJobs: one declared owner per job, a revision-pinned switch, a rollback that could not be a revert, and alerts that had to move too.8 min
  10. 10Helmsman: when a research candidate becomes infrastructureThe design of an explicit promotion boundary: frozen definitions, concurrent requests, durable evidence gates, and the limits of a deployed research workspace.10 min