Applied AI and quantum research, tested against engineering evidence
Our innovation agenda is simple: find problems worth testing, compare the established method with the emerging one on real data, and report the result whether or not it flatters the new idea. Applied AI and quantum research run as one program, staffed by the engineers who will have to build whatever survives.
Research and experimentation
Three strands, each with current work we can name without naming clients.
Applied AI research
Current threads: multi-model document extraction, where several vision-language models read the same invoice and a judge model selects the best result; intent-classification benchmarking for conversational assistants, comparing LLM and retrieval approaches on the same utterances; and lightweight edge inference for cameras and gateways. Each thread ends in a benchmark note and, where it works, a component in a product.
Quantum research
Algorithms, simulation, hybrid quantum-classical workflows and cryptographic readiness, with current limitations documented alongside the results. Our physics-aware validation framework, a research-in-progress method that checks candidate solutions against domain constraints before expensive computation, is the thread furthest along. Client quantum deployments do not exist yet, and the page will say so until they do. Read the quantum service page for the six research areas.
Innovation and proof-of-concept lab
Scoped experiments that assess data readiness, technical fit, business value, risk and the conditions for wider delivery, typically in 4 to 8 weeks on your data. An experiment is a decision tool: it tells you whether to fund the next step. It is never a production guarantee, and we write the report so that a no-go is as useful as a go.
How does an innovation challenge run?
You bring a problem; we frame it as a testable question. The steps mirror our delivery approach but stop after the prototype decision.
- 01
Frame the question
A one-page hypothesis: what should improve, by what measure, against which baseline.
- 02
Secure the data
Representative samples under NDA, hosted where your residency rules require.
- 03
Baseline the classical method
The incumbent approach is measured first so the new one has something to beat.
- 04
Run the experiment
Time-boxed build, measured against the hypothesis, limits recorded.
- 05
Decide
A written go, no-go or reshape recommendation, with the engineering path if it is a go.
Research with a deadline attached
Some of our research is paced by regulators rather than curiosity. Post-quantum cryptography readiness is the clearest case: NIST published FIPS 203 and 204 in August 2024, SAMA's 2026 circular asks Saudi banks for a cryptographic inventory by 31 December 2026, the UAE National Encryption Policy of November 2025 asks government entities for transition plans, and India's National Quantum Mission task force has critical sectors starting PQC implementation by 2027. Our PQC readiness work turns that research into inventories and migration plans, and it is the part of our quantum program clients can use today.
Propose an innovation challenge
Bring a problem worth testing. The best candidates have a measurable baseline, data you can share under NDA and a sponsor who wants an honest answer. We respond with a framed hypothesis and a scoped experiment plan, and we will tell you if the question is better answered by a production engagement than by research.
Frequently asked questions
What does an innovation challenge cost and how long does it take?
Experiments are fixed-price and time-boxed, usually 4 to 8 weeks depending on data preparation. The price covers framing, baseline, experiment and the decision report. If the outcome is a go, the report includes an estimate for the engineering path so you can budget the next step without a second discovery phase.
Who owns the results and any intellectual property?
Results produced on your data belong to you, including the decision report and any prototype code built specifically for the challenge. Our general methods, frameworks and pre-existing components remain ours and are licensed for your use. We agree this in the statement of work before any data is shared, and we are happy to work under your standard research agreement.
Can you run a quantum experiment for us today?
Yes, as research. We can frame an optimization or simulation question, build the classical baseline, run quantum-inspired and simulated quantum methods against it and report the comparison with the hardware limits stated. What we cannot do is promise a quantum advantage, and we have not yet deployed a quantum method for a client. The report will say which of those categories your problem falls into.
Do you publish your research?
We publish method notes and benchmark summaries where clients permit and where the result is useful to others, and we document limitations in the same note. Client data and identities are never published. Our aim is credibility through method rather than announcements, which is also why this page names research threads but not clients.
Bring a problem worth testing
Describe the question, the baseline and the data you have. We will frame the hypothesis and propose a scoped experiment within two weeks.