Work
The work speaks quietly.
We measure success by what changes inside an organization: research that reaches an answer sooner, systems that hold under pressure, and teams that see the whole picture. Below is how we built it.
Technology
- Rust
- Tauri
- TypeScript
- React
- Python
- vectorbt
- numba
- CodeMirror
- xterm.js
- WebRTC
- Cloudflare Workers
- Durable Objects
- MongoDB
- Stripe
- Claude
- MCP
- Interactive Brokers
- OANDA
Finance · Digital Products · QuantIDE
QuantIDE: an institutional research desk, built for the individual
Quantitative tools were built for institutions: expensive, fragmented and closed. Independent traders stitched together editors, notebooks, data feeds and broker terminals, and lost confidence in every result along the way.
A native workstation in Rust and React
QuantIDE is a Tauri desktop application with a Rust core and a React and TypeScript interface. The core manages real terminals, Python execution, file watching, Git and process supervision; the interface carries a CodeMirror editor, WebGL-accelerated terminals and live charting. Signed builds ship for macOS, Windows and Linux through an automated release pipeline with in-place updates.
Research that streams as it runs
Backtests stream their equity curve and drawdown to the chart while the engine walks the bars, so a trader sees a strategy succeed or fail as it happens. Parameter sweeps can be re-run from interface controls, and research builds on the Python ecosystem professionals use, including vectorbt, numba and Parquet.
An AI engineer with real tools
The built-in assistant runs on Claude and works through Model Context Protocol tools: it writes strategies, runs backtests and renders results as live charts and tables. Tool results register as datasets, so large data never has to pass back through the model. Usage is metered per user at the edge, and traders can bring their own provider if they prefer.
From research to the broker
Strategies connect to Interactive Brokers, OANDA and MetaTrader. Long-running trading programs run under a supervisor process that survives crashes and closed windows, stream fills, signals and equity to a live desk, and refuse to trade a live account unless explicitly allowed.
Cloud services at the edge
Accounts, billing, AI metering, rooms, messaging and a live lecture hall run on Cloudflare Workers and Durable Objects, with WebRTC media through a relay and signalling sharded across instances. The platform API runs on Next.js with MongoDB and Stripe.
Finance · Platforms & Portals · A multi-generational family office
Algorithmic trading, from research to live execution
Strategies were researched in one place, tested in another and traded by hand, so no one could say with confidence that what went live was what had been tested.
One strategy, three environments
Strategies are written once against an event-driven interface and run unchanged in backtest, paper and live trading. Historical data is replayed through the same order path the broker adapters use, which removes the most common source of difference between research and reality.
Data the research can trust
Point-in-time histories adjusted for corporate actions, survivorship-free universes and partitioned columnar storage, queried directly from research code. Every backtest records the data snapshot, code version and parameters that produced it.
Execution and pre-trade risk
An order management layer enforces position, notional and loss limits before any order leaves the building, slices large orders over time, and keeps an immutable record of every state an order passes through. A single control halts all trading.
Reporting for principals
Daily P&L, exposure and attribution by strategy, asset class and currency, consolidated across accounts and delivered through a private, permissioned dashboard.
Finance · Automation & AI · A quantitative hedge fund
A research platform that lets the analysts do research
Analysts spent more of their week cleaning data and maintaining scripts than testing ideas, and results were hard to reproduce from one researcher to the next.
Pipelines with contracts
Orchestrated ingestion of market, fundamental and alternative data, each feed validated against a schema and statistical checks before it reaches research. Datasets are versioned, so any past result can be re-run on exactly the data it saw.
A shared research library
A common library of signals, factor definitions, transaction-cost models and vectorized backtests, compiled with numba where speed matters, so every analyst builds on the same tested foundations.
Compute on demand
Parameter sweeps, walk-forward validation and model training fan out across a cluster and scale back to zero when idle, turning overnight jobs into minutes.
Machine learning and language models
Models trained on alternative data with strict separation of training and test periods, and language models that extract structured signals from filings and transcripts, each extraction reviewable by a person.
Finance · Technology Advisory · A leading US bank
Financial engineering at the scale of a national bank
Critical analysis depended on tools that had grown up around individual teams, at a scale and under a level of regulatory scrutiny where errors are not an option.
Pricing and risk analytics
Libraries for valuation, sensitivities and scenario analysis, vectorized or compiled where performance matters, and reconciled against the bank's systems of record before anyone relied on them.
Built for model risk review
Every model documented, versioned and reproducible in the form model risk management expects: assumptions, limitations, test results and independent validation in one place.
Data lineage and controls
Every figure traceable to its source, with entitlements and audit logging that satisfy internal audit and the regulators behind it.
Batch and grid computation
Large scenario and stress runs distributed across compute and scheduled to finish before the business day begins.
Enterprise · Automation & AI · A global technology and retail company
Engineering inside one of the world's largest operations
At global scale, small inefficiencies are multiplied millions of times, and any new system has to fit an estate already running at full speed.
Event-driven and serverless
Services built from functions, queues, workflows and managed databases that scale with demand and cost nothing when idle, with every resource defined as infrastructure-as-code.
Automation that fails safely
Orchestrated workflows with retries, idempotent steps and dead-letter handling, so a failure is isolated, visible and recoverable rather than silent.
Operational excellence
Metrics, alarms, canaries and dashboards from the first deployment, staged rollouts with automatic rollback, and runbooks reviewed before go-live.
Defense · Platforms & Portals · The US Army and US Air Force
Systems built to the standard of the armed forces
Defense work leaves no room for fragile software: systems must be secure, dependable and usable by people under pressure.
Security by construction
Hardened container images, CAC and PKI authentication, role-based access, encryption in transit and at rest with validated cryptography, and an audit record of every action.
Ready for accreditation
Controls implemented and documented against the Risk Management Framework from the start, so the path to authority to operate is evidence, not retrofit.
Dependable where it is used
Delivered to government cloud and on-premises environments with repeatable, infrastructure-as-code builds, and designed to degrade gracefully when networks are slow or unavailable.
Construction · Platforms & Portals · A construction and project management firm
Every project, from bid to handover, in one place
Schedules, budgets, field reports and client updates lived in spreadsheets, inboxes and site binders, so the true state of a project was always a few days old.
A schedule that recalculates itself
Tasks, dependencies and durations feed a critical-path engine, so a slipped delivery immediately shows which milestones move and by how much.
Built for the field
An offline-first mobile app for daily logs, photos, inspections and punch lists, time- and location-stamped and synchronised without conflicts when signal returns.
Cost and documents connected
Budgets, commitments and change orders reconcile with the accounting system, and drawings, RFIs and submittals carry full version history and approval trails.
We name clients only with their permission.