tracord.org/technology

Built for calculation-heavy finance

A small, deliberate stack: a Go calculation core, Python analytics services, TypeScript clients, and a data layer built for market streams and large portfolio jobs.

01 / Languages

Each language has one job

Go
Calculation core
  • Pricing, yield and duration from cash flows
  • Portfolio and risk jobs over large books
  • Market-data processing and scanners
Python
Analytics services
  • Returns, contribution and attribution
  • Analytics APIs on FastAPI
  • Reporting and data-quality logic
TypeScript
Client applications
  • Web clients on Angular
  • Dashboards and interactive charts
  • gRPC-Web and REST to the backend
02 / High-load path

Streams in, numbers out

Market data arrives as a stream. Portfolio calculations then run across holdings, prices and cash flows, often for many portfolios on the same day.

PostgreSQL holds the operational data. Redis keeps hot reads close to the services. ClickHouse stores the analytical series we scan across time. Services talk over gRPC, and queues take the work that should not block a request.

Everything runs in containers on Kubernetes, so a calculation job and an API scale on their own.

stack.layers
01ingest
KafkaSQS-compatible queuesFAST feeds
02store
PostgreSQLRedisClickHouseS3
03compute
GoPython
04serve
gRPCgRPC-WebRESTFastAPI
05clients
TypeScriptAngular
06run
KubernetesDockerHelmGitLab CI
07observe
PrometheusSentry
08access
OpenID ConnectJWTroles
03 / Engineering practice

Numbers people can rely on

Financial software is judged by whether the figure is right. Most of our engineering effort goes into keeping it right as the code changes.

Tested calculations

Calculation results are checked against frozen reference snapshots, so a refactor cannot quietly move a number.

Reproducible delivery

Vendored dependencies, container images and CI pipelines from commit to staging to production.

Observability

Metrics, error tracking, slow-query logs and event-loop stall detection on live services.

Access control

Single sign-on over OpenID Connect, token-based APIs and role-based permissions.

Next step

Have a workflow that should run by itself?

Tell us how the process runs today: sources, spreadsheets, reports. We will come back with what can move into software and what it would take.