AI & Platform Architect · Regulated Industries

I build AI for industries where “close enough” isn’t.

Life sciences. Tax & finance. Digital assets. I design the self-hosted LLMs, document AI and cloud platforms underneath, and I measure them on real hardware before anyone relies on them.

  • 3regulated domains
  • 100+page documents, read by AI
  • 0guesses: measured or shown
~/gpu-fit
$ pip install gpu-fit
$ gpu-fit Qwen/Qwen3-8B --gpu L40S --context 32k

Qwen/Qwen3-8B  on  1× L40S (45.0 GiB)
  vLLM budget per GPU          40.5 GiB
    weights                    15.3 GiB
    runtime overhead            1.2 GiB
    left for KV cache          24.0 GiB
  KV cache capacity       174,639 tokens
  Requests at full context        5.3

✓ Fits. Suggested command:
  vllm serve Qwen/Qwen3-8B --max-model-len 32768
Before you rent the GPU: gpu-fit

Where I work

One thread: technology that has to be right.

Life sciences

Enterprise architecture and technology leadership in global pharma, where systems are validated and every change is traceable.

Tax & finance

AI platforms for transfer pricing and tax compliance: reading financial statements, benchmarking, drafting reports used for real filings.

AI in production

Self-hosted LLMs on GPUs with vLLM, long-document AI, and gateways that route between cloud and private models with per-call cost tracking.

Digital assets

Wallets and token infrastructure, built with the same discipline: keys, ledgers and edge cases first.

How I work

  1. 01

    Measure, then claim.

    Every number I publish comes from a run on real hardware or arithmetic you can check. Estimates are labelled.

  2. 02

    Deterministic where it counts.

    Money, ranges and percentiles are code with tests. AI does the reading and drafting, never the maths.

  3. 03

    Private by design.

    When data can’t leave, the model comes to the data: open models on your own GPUs, sized properly.

  4. 04

    Lead, and still ship.

    I run teams and architecture, and I still write the code, run the benchmarks and read the logs.

Open source

gpu-fit

How many users can this LLM serve on this GPU? Most calculators multiply parameters by two. gpu-fit reads the model’s real architecture (grouped-query, hybrid, MLA, sliding-window, FP8) and mirrors vLLM’s memory budget. Then it hands you the command.

  • Validated against real vLLM runs on an L40S
  • Warns before vLLM refuses to start a hybrid model
  • No dependencies · MIT licence · Python 3.10+
GitHub →
Grouped-query attentionHybrid linear attentionDeepSeek MLASliding windowFP8 / AWQ / GPTQTensor parallelFP8 KV cacheVision encoders

Building something where accuracy isn’t optional?

I’m always glad to compare notes on private LLMs, document AI or regulated platforms.

Find me on GitHub