Make it work
We'll make it faster

Numerical code optimisation as a service.
Tailored for scientific computing and quantitative finance teams.

Real optimisations on public repositories

Verified improvements on functions from open source numerical Python projects. Correctness checked end-to-end.

skfem
_adaptive()
skfem/mesh/mesh_tet_1/MeshTet1.py
−93.9%
runtime reduction
Adaptive tetrahedral mesh refinement optimised for finite element solvers. 116/116 tests passing, correctness verified.
skfem
split()
skfem/assembly/basis/composite_basis/CompositeBasis.py
−79.5%
runtime reduction
Composite basis splitting streamlined to avoid intermediate NumPy allocations. 1/1 tests passing, correctness verified.
skfem
_adaptive_split_elements()
skfem/mesh/mesh_tri_1.py
−69.1%
runtime reduction
Adaptive triangular mesh element splitting rewritten with preallocated output arrays. 11/11 tests passing, correctness verified.
skfem
_adaptive_sort_mesh()
skfem/mesh/mesh_tri_1.py
−63.9%
runtime reduction
Adaptive triangular mesh sorting rewritten with in-place difference arrays and squared lengths. 11/11 tests passing, correctness verified.
skfem
_adaptive_sort_mesh()
skfem/mesh/mesh_tet_1/MeshTet1.py
−46.9%
runtime reduction
Adaptive tetrahedral mesh sorting rewritten with batched edge-length computations. 470/470 tests passing, correctness verified.
skfem
_intersect2d()
skfem/supermeshing.py
−20.4%
runtime reduction
2D supermesh intersection routine used in non-matching mesh transfer operations. 1/1 tests passing, correctness verified.

Built for codebases where correctness matters

Three principles guide every optimisation DeltaCode delivers.

Tailored objectives

Optimise for what matters

Target runtime, memory consumption, lines of code, or readability depending on your codebase priorities.

Secure environment

Isolated containers

Code runs in isolated containers configured for your specific environment, with real execution data and verified outputs.

Controlled AI

Deterministic workflow

LLM output is benchmarked and validated against the original implementation before any pull request is opened.

Two ways to work with DeltaCode

Choose a managed engagement with our team, or join the waitlist for self-serve access at launch.

For businesses
Managed service
We onboard your codebase and configure entry points
Optimisation runs are managed end to end in secure containers
Results delivered as GitHub PRs, correctness verified, with auditable documentation
For codebases where runtime matters
Custom pricing per engagement
SaaS platform
Join the beta waitlist
Self-serve GitHub integration, connect in minutes
Automated scanning on every push or on a schedule
Dashboard to monitor PRs, results and scan history
Currently in private beta, limited spots available
Per-seat pricing at launch

Frequently asked questions

What languages do you support? +
DeltaCode currently supports numerical Python codebases. Support for additional languages, including C++ and Julia, is planned but not available yet.
How does correctness verification work? +
Every optimised function is executed end-to-end inside a secure container against your actual entry scripts. The optimised version must produce identical outputs to the original for every collected test case before a pull request is opened. If correctness verification fails, no PR is created.
What data does DeltaCode store? +
DeltaCode stores function-level optimisation results, runtime metrics, test outputs, and pull request metadata to enable history tracking and reporting. Your full source code is not retained beyond the duration of an active optimisation run. Encrypted GitHub tokens are stored to enable webhook integration.
What kind of codebases benefit most? +
Compute-heavy numerical Python codebases benefit most — scientific simulations, quantitative finance models, finite element solvers, signal processing pipelines, and data analysis workloads. DeltaCode is built for code where runtime matters and correctness is non-negotiable.
How is pricing structured? +
The managed service is priced per engagement based on codebase complexity, number of functions, and scope of work. The SaaS platform will be priced on a per-seat monthly basis when it launches publicly. Contact the team for current pricing and details specific to your situation.
When will the SaaS platform be available? +
The SaaS platform is currently in private beta with a limited number of users. Public launch is planned for 2026. Join the waitlist to be notified when a spot becomes available.