Engagement

Engagement structure, pricing, research-data handling, and initial-contact guidance.

Engagement Formats

Fixed-scope project
Scoped, milestone-paid

A scope defined and agreed in writing, with milestone-based delivery and payment. The most common format; appropriate for paper reproduction, client-research implementation, productionization, and demos.

Hourly consulting hours
Block of hours

A purchased block of engineering or research-engineering hours, drawn down over an agreed period. Appropriate for benchmarking, code review, and short investigations.

Paper reproduction quote
Per-paper flat

A flat quote for reproducing a specific public paper end-to-end. The quote is issued following a complimentary preliminary review of the paper and any official artefacts.

Scope and Duration

Most engagements are completed within a one-to-eight-week period. We undertake a single primary engagement at a time, with at most one short secondary engagement in parallel. We do not undertake engagements exceeding approximately three months under a single contract; for work of larger scope, we propose dividing it into a sequence of phased engagements, each concluding with an independent continuation decision.

Pricing

Pricing depends on the engagement format. Fixed-scope projects are priced from estimated effort, with an additional buffer for components of genuine uncertainty. Hourly consulting hours are sold in a minimum block and billed at a fixed hourly rate. Per-paper reproduction quotes are bracketed by paper size and difficulty. In every case, the preliminary assessment that produces the quote is provided free of charge, and the price is communicated before any work commences.

Included

  • Review the materials in full
  • Communicate whether the engagement is within our scope
  • Provide a defined price and timeline

Excluded

  • Quote without reviewing
  • Undertake an engagement we do not believe we can complete
  • Quote ambiguous ranges and revise subsequently

IP & Confidentiality

By default, the source code, documentation, and reports we deliver are assigned to the client. Open-source dependencies we incorporate remain under their respective licenses, which are listed in the deliverable. An NDA will be executed prior to the exchange of any sensitive material — see §8.6 for the specific handling of unpublished research. We do not claim authorship credit on any algorithmic contribution; the method itself remains the client's intellectual property. Should the client elect to release the resulting code as open source alongside a publication, we will support the chosen release process.

Data and Unpublished Research

Most of the following applies to every engagement; the second subsection applies specifically to unpublished papers, drafts, and unsubmitted methods, which is the case for the majority of our research-side clients.

1. Working Environment

By default, we operate on client-provided machines or within the client's cloud account, using credentials that the client provisions and can revoke. When local work is unavoidable, we use an isolated, encrypted working directory dedicated to the engagement, with no cross-mounting from other projects. We do not archive client datasets or retain training-set copies for contingency purposes; any local data is transient working state, not a personal repository.

2. Unpublished Research

  1. NDA executed prior to exchange of sensitive materials. We execute the NDA before the client transmits the draft, derivation, or method description. The initial communication may describe the general nature of the problem; specific details are exchanged once the NDA is in effect.
  2. Drafts and method descriptions are treated as client confidential IP. The draft, the derivation, the method description, and the resulting code are all treated as confidential material, handled in accordance with the same protocols as any client confidential material.
  3. No overlapping engagements in the same research domain. While we are implementing the client's method, the assigned engineer does not undertake another engagement in a closely related research domain where the methodologies could plausibly intersect.
  4. No public-channel debugging involving the client's algorithm. We do not post code or descriptions containing the client's unpublished algorithm to public channels — including Stack Overflow, public GitHub issues, public forums, or consumer LLM chat services. Generic technical questions (e.g., regarding CUDA errors) may be raised in public channels; the algorithm itself is never disclosed.
  5. Third-party AI tools require prior written approval. Should a third-party AI tool be required during implementation, the set of approved tools and the boundary of permitted data transmission are agreed in writing prior to use. By default, consumer LLM products involving data upload are not used.
  6. Working copies stored in agreed locations only. Working copies are stored on encrypted local disks or in a private git repository specified by the client. They are not stored in personal cloud drives, personal note-taking applications, or shared workspaces.
  7. Post-engagement deletion upon request. Upon delivery, the client may instruct us to delete all working copies. We retain only the artefacts explicitly approved for retention by the client (typically: the final deliverable repository and the NDA-bound implementation report).
  8. The engagement is not publicised by us. The engagement is not presented as a case study on our website, in proposals submitted to other clients, or in any external material, regardless of whether the paper is subsequently published, accepted, rejected, or withdrawn.
  9. Confidentiality obligations survive publication. Our confidentiality obligations continue after submission, acceptance, rejection, or withdrawal of the paper, for the duration specified in the NDA.
  10. No authorship claim. We do not claim authorship credit on any algorithmic contribution. The method itself, and any novelty therein, remain the client's intellectual property.

3. Published-Paper Reproduction

For engagements involving the reproduction of a publicly available paper, the method itself constitutes public knowledge, and the implementation may typically be open-sourced or released alongside the client's follow-up work. However, any subsequent work the client's team develops based on the reproduction — datasets, new method variants, internal experiment results — is treated under section 2 above in the same manner as any other unpublished research.

4. Post-Delivery

Four weeks after delivery, we delete all working copies by default — local repositories, log directories, intermediate datasets, and exploratory notebooks. Should the client wish for us to retain copies for potential follow-up maintenance, please notify us in writing; we will retain precisely the items specified, on the storage location specified.

Languages & Response Times

Initial Contact

An initial inquiry by email is sufficient. For work involving unpublished research, please request an NDA in the initial communication prior to transmitting any specific details — we will prepare and execute the NDA expeditiously. Otherwise, the more concrete the initial inquiry, the more efficiently we can provide a useful response (acceptance, decline, or a request for additional information).

Please describe the algorithm, paper, or prototype for which assistance is sought. A brief paragraph and a relevant link are sufficient to initiate the discussion.

For unpublished research, an NDA may be requested in the initial email prior to transmitting specific details.