Quantum / Classical
Comparative evaluation
Plan scoped comparisons of classical, quantum and hybrid methods using a defined dataset, baseline and cost model. Deliver reproducible experiments and a report that explains the result.
TranscendQ LLC · Quantum computing research & engineering
Quantum computing,
tested.
We develop reproducible evaluation tools for quantum and classical computing, with an initial focus on quantum error-correction decoders and GPU–FPGA engineering.
Discuss a projectFounded in Texas in October 2026 by Geonhee Cho. Our decoder evaluation platform is in development. University-origin technology is commercialized only under the applicable licensing and agreements.
Research and technical work
Define the question. Build the test. Inspect the evidence.
QEC / Hardware
Decoder engineering
Develop evaluation workflows that connect an error-correction decoder's mathematical reference with its hardware implementation. Record output agreement, fault handling and latency separately.
Research / Software
Reproduction and tooling
Turn a research method into a repeatable software experiment with documented inputs, configuration records and test scripts. Project scope and acceptance criteria are agreed before work begins.
The proposed product
A traceable path from configuration to result.
Decoder engineers need to distinguish an algorithm change from a hardware change—and preserve the evidence needed to investigate a mismatch.
Declared configuration → Reference workload → Implementation execution → Comparison report
01 / Reproduce
Configuration manifests
Record graph, weight, workload and implementation identities so an evaluation can be repeated within a declared scope.
02 / Compare
Reference-first checks
Compare implementation output against a specified software reference. GPU-assisted batch checking is planned; it is not claimed to be a GPU decoder.
03 / Investigate
Actionable validation reports
Separate agreement, quality, throughput and end-to-end latency. Preserve raw records and explain what each measurement does—and does not—establish.
Evidence policy
Claims stay inside the test boundary.
We distinguish simulation, physical execution and customer evaluation. Detailed non-public artifacts are shared only after release and confidentiality review.
Current development status
Work is focused on integer-weighted union–find decoder evaluation and reproducible GPU–FPGA testing. Existing research artifacts are being organized into a validation package. Product-level host-input processing, sustained workload testing and customer evaluation remain development milestones.
What is not claimed
No production readiness, customer adoption, QPU-integrated latency, experimentally established threshold, universal correctness or certified worst-case end-to-end latency is claimed. Timing and correctness results apply only to their documented configuration and measurement conditions.
What a pilot would evaluate
A jointly defined workload, baseline, configuration-change scenario and acceptance criteria. A pilot would compare reproducibility, output agreement, fault reporting and measurement overhead—not presume a performance advantage.
Development roadmap
From research artifacts to an evaluation tool.
Milestone A
Independent reproduction
Verify artifacts, identities and reference comparisons; distinguish simulated records from physical board captures.
Milestone B
Live input and fault handling
Develop host-to-FPGA framing, configuration identity checks, negative tests and sustained-input measurements.
Milestone C
Scoped pilot evaluation
Extend workloads and establish a customer-agreed protocol with explicit metrics, limitations and deliverables.
Try the development prototype
A decoder log. A check. A report you can inspect.
Load a synthetic execution log or paste non-confidential JSONL. Inspect configuration, correction and syndrome comparisons, with record-level evidence citations.
Open the decoder log demo →The current demo runs a deterministic checker locally in your browser. Claude API integration is prepared separately and is not connected. Synthetic examples do not represent customer or QPU results.
How we build
AI-assisted development. Independently checked results.
We plan to use Claude for software development, test preparation, experiment-log analysis and report drafting. Mathematical checks, reference implementations and repeatable experiments provide the evidence used to assess each result.
Our goal is a practical workflow that helps engineers investigate decoder changes and researchers compare computing methods within a declared scope.
Start a technical conversation
Tell us what you need to evaluate.
Founder: Geonhee Cho, PhD
gunheecho1221@gmail.com
For an initial discussion, share the problem, available data, computing environment and decision the evaluation should support.
Please do not send proprietary workloads or confidential implementation details before a suitable agreement is in place.