Tomas
Jimenez.
Principal Software Engineer
I build C++ low latency systems, IBM mainframe software, compilers and GPU/ML tooling.
Now exploring
deterministic-llm-compilerAbout me
I'm an engineer who believes fast systems and clear thinking belong together.
I work on C++ low latency systems, IBM mainframe software, compilers, CUDA and ML. This site is where I write things down: essays on C++, CPUs and systems, and research on compiler-validated LLM code generation and real-time physics simulation.
02
Research papers
- 02
- Published
- 03
- Posts
- 17
- Topics covered
- μs
- Unit I think in
What I do
Low latency
Cache lines, memory layout, and shaving microseconds.
Compilers
Intermediate representations, code generation and compiler-validated LLM output.
Mainframe
IBM z/OS, Db2, CICS, IMS, RACF, JCL, Rexx, COBOL and z/Architecture Assembler.
Research
Compilers, LLM code generation, numerical methods and physics engines.
Writing
Essays on C++, CPUs and the systems in between.
Skills
Tech
stack
Languages
- C++20
- C
- Python
- Java
- COBOL
- z/Architecture Assembler
Systems
- Low latency
- Concurrency
- Linux and Windows internals
- IBM z/OS
- Db2
Accelerators and ML
- CUDA
- LLM fine tuning
- Compilers
Cloud
- Azure
- AWS
- GCP
Latest writing
All posts
C++ is not dying in quant finance - it's getting more critical
Sep 29, 2026 · c++, quant, low-latency
- Inside the CPU: how the processor talks to everything else
Inside the CPU: how the processor talks to everything else
Sep 29, 2026 · cpu, hardware, low-latency, youtube
- Modern C++ Deep Dives: a new YouTube series
Modern C++ Deep Dives: a new YouTube series
Sep 29, 2026 · c++, youtube, data-structures
Research
All research- 01
Improving LLM Code Generation Reliability Through Compiler-Validated Intermediate Representations
A controlled study protocol testing whether a typed intermediate representation, deterministic compiler validation and one bounded repair attempt make LLM-generated code more reliable. Using the Seq language's compiler front end and C back end, five comparison arms separate the effects of the IR, grammar-constrained decoding and compiler-guided repair across 120 tasks, with functional correctness on independent tests as the primary outcome. No experimental results are claimed yet.
- 02
minicollide: Numerical Methods for Real-Time Rigid-Body Dynamics in a Compact, Deterministic Physics Engine
A from-scratch 3D rigid-body physics engine built as a laboratory for the numerical methods behind real-time simulation: symplectic integration, sweep-and-prune broadphase, SAT and GJK+EPA collision detection, and a warm-started sequential-impulse solver. Constraint islands are solved in parallel with bit-exact determinism, validated with four instrumented demos.
From the blog
Latency doesn't care about your productivity argument.
Let's talk about something
interesting.
Hiring for a Principal or Staff role in C++, low latency systems, IBM mainframe, compilers or GPU/ML? Or a question about a post or a paper? Message me on LinkedIn.