Skip to content
Berke Tok
Back to work

Capital markets

Kestrel Capital

Research to execution, on one rail

Backtest, blotter and exposure in one terminal, running the code that will trade it
Sector
Capital markets
Location
DIFC, Dubai
Year
2025
Kickoff to production
22 weeks
01Overview

Backtesting, execution and risk for a DIFC fund, rebuilt so a strategy moves from notebook to live capital in a day.

02The problem

Strategies were researched in notebooks and re-implemented by hand for production. The two versions never behaved identically, so every discrepancy became a week of forensics. Risk limits lived in a spreadsheet.

Risk and blotter on one desk: same code that ran the backtest
03What we built
  1. 01

    One codebase, two modes

    The same strategy code runs against historical and live data. Simulation models real fees, funding and market impact, so backtests stop flattering themselves.

  2. 02

    Risk before the order

    Position limits, exposure caps and kill switches evaluate pre-trade. Breaches halt the strategy and page a human rather than logging a warning.

  3. 03

    Reconciled every morning

    Fills, fees and positions reconcile against venue statements automatically. Any break is flagged before the desk opens.

A strategy reaching live capital only once every pre-trade limit has passed
Pre-trade kill switch paging the desk before the order leaves
04The result

Research-to-production now fits in a day, and backtest-to-live tracking error is small enough to be a metric instead of an argument.

What changed

Research to live
1 day
Reconciliation
Daily
Risk limits pre-trade
Live
Fills, fees and positions reconciled overnight, with both breaks named before the open

Stack

  • Python
  • Rust
  • TimescaleDB
  • Redis
  • Grafana

Disciplines applied

  • Financial technology
  • Data & analytics
  • Custom software development

Closing

Do you have an idea, or a problem?

Either works. In my experience the second one goes faster.

I usually reply the same day.