This weekends work was a machine learning process pipeline that blends classical analysis, local LLMs, Anthropic Claude Code, and IBM Quantum.
The automated workflow is fairly simple in concept:
- ๐ฆ๐๐ฎ๐๐ถ๐๐๐ถ๐ฐ๐ฎ๐น ๐ฝ๐ฟ๐ผ๐ณ๐ถ๐น๐ฒ๐ฟ benchmark the data locally using classical methods.
- ๐ฆ๐๐ฟ๐ฎ๐๐ฒ๐ด๐ถ๐ฐ ๐ฑ๐ฒ๐ฐ๐ถ๐๐ถ๐ผ๐ป ๐ฒ๐ป๐ด๐ถ๐ป๐ฒ running locally via Ollama (llama3.1:8b) to decide whether we should proceed with a Quantum solution, if suitable then proceed with the next steps.
- ๐๐ผ๐ฑ๐ฒ ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐ to generate the IBM Qiskit code using Claude only if the case is strong enough
- ๐ค๐๐ฎ๐ป๐๐๐บ ๐ฒ๐ ๐ฒ๐ฐ๐๐๐ผ๐ฟ to handle submission to IBM Quantum hardware
The repo is public (๐ญ๐ช๐ฏ๐ฌ ๐ช๐ฏ ๐ต๐ฉ๐ฆ ๐ค๐ฐ๐ฎ๐ฎ๐ฆ๐ฏ๐ต๐ด) and the point of the exercise was not to chase quantum for the sake of it, it was to explore a few practical questions:
- How accessible is the Qiskit and the quantum world when using tools like Claude Code?
- How much of circuit development and submission can be automated?
- How far can I push a workflow that starts with classical benchmarking and only moves to quantum when there is at least some case for doing so?
- โฆ and most importantly learn!
Iโm really happy with how this turned out. I got the circuit and jobs built and submitted successfully. They do seem to time out, which may simply be a limitation of the open plan, but even with that, I came away extremely positive about the result.
Couple of takeaways:
- GenAI makes the quantum world feel more accessible.
- Not mature. Not simple. Not yet industrialised. But more accessible.
- Claude Code was able to help bridge some of the gap between imagination and implementation - cost $0.5c for the code generation. (๐๐ฉ๐ช๐ค๐ฉ ๐ฎ๐ข๐ต๐ต๐ฆ๐ณ๐ด, ๐ฃ๐ฆ๐ค๐ข๐ถ๐ด๐ฆ ๐ฐ๐ฏ๐ฆ ๐ฐ๐ง ๐ต๐ฉ๐ฆ ๐ฃ๐ข๐ณ๐ณ๐ช๐ฆ๐ณ๐ด ๐ต๐ฐ ๐ธ๐ฐ๐ณ๐ฌ๐ช๐ฏ๐จ ๐ช๐ฏ ๐ฏ๐ฆ๐ธ๐ฆ๐ณ ๐ต๐ฆ๐ค๐ฉ๐ฏ๐ช๐ค๐ข๐ญ ๐ฅ๐ฐ๐ฎ๐ข๐ช๐ฏ๐ด ๐ช๐ด ๐ฐ๐ง๐ต๐ฆ๐ฏ ๐ฏ๐ฐ๐ต ๐ซ๐ถ๐ด๐ต ๐ต๐ฉ๐ฆ ๐ค๐ฐ๐ฏ๐ค๐ฆ๐ฑ๐ต๐ด, ๐ฃ๐ถ๐ต ๐ต๐ฉ๐ฆ ๐ง๐ณ๐ช๐ค๐ต๐ช๐ฐ๐ฏ ๐ฐ๐ง ๐ต๐ฐ๐ฐ๐ญ๐ช๐ฏ๐จ, ๐ด๐บ๐ฏ๐ต๐ข๐น, ๐ข๐ฏ๐ฅ ๐ฑ๐ณ๐ฐ๐ค๐ฆ๐ด๐ด.)
For me, the real insight from todayโs work - not proving that quantum is ready for every day use, but proving that the path to testing, learning, and building with it is getting a lot more approachable (๐ข๐ฏ๐ฅ ๐ข ๐ญ๐ฐ๐ต, ๐ญ๐ฐ๐ต ๐ฒ๐ถ๐ช๐ค๐ฌ๐ฆ๐ณ).
GitHub Repo: https://github.com/steveh250/IBMQuantum-Agent