Design judgment from senior engineers.
Coding models can write a function. Designing a system that survives production is harder. We commission senior engineers to create and grade the data that teaches it.
Software architecture, with proof.
Code models produce designs that read well and fail under load: a single database with no plan for growth, retries without idempotency, a cache with no invalidation strategy, secrets in configuration, or a migration that cannot be rolled back. Agents make similar mistakes over longer horizons. HUMXN commissions credential-checked staff and principal engineers, SREs and security engineers to write system designs, review code and grade agent trajectories against the standard a senior reviewer would apply.
Every item is written to your specification, fingerprinted at intake and checked for duplicates, then reviewed by a second senior engineer. Code and documents are scanned for personal information and masked, and label provenance records whether each judgment came from the author, a model suggestion or an expert reviewer. Deliveries come with a signed receipt and audit trail, in JSONL, Parquet or Croissant.
For senior engineers, this is paid, remote work that uses the judgment you exercise in design reviews and incident retrospectives. You write design documents, review code and agent runs, and check a colleague's work, at a rate shown before you accept each job. Work fits around a full-time role, subject to your employer's policies. Your identity is not published, and approved work is paid automatically through Stripe.
Built to your specification.
Every engagement starts from a written spec and a pilot batch. These are the most common requests we source for software architecture.
System design documents
Design docs for realistic problems, with requirements, trade-offs, failure modes and the alternatives that were rejected.
Architecture decision records
Expert-written decisions with context and consequences, used to teach models how engineers reason about trade-offs.
Code review data
Pull requests reviewed for correctness, security, performance and maintainability, with comments a senior reviewer would leave.
Agent trajectory grading
Engineers grade multi-step coding agent runs, marking where a plan went wrong and what a better next step would be.
Preference judgments
Comparisons of model designs and code on correctness, simplicity and operational risk, with written rationales.
Design evaluation sets
Held-out design prompts with rubrics and reference answers, newly written to avoid public interview-question contamination.
01Plausible but fragile designs
Designs that ignore failure, scale and operations look fine on paper. Senior engineers judge them as they would in review.
02Long-horizon agent errors
Agents drift over many steps. Step-level grading locates the decision that sent a run off course.
03Licensed code
Scraped repositories carry unclear licenses and secrets. Consent bound to each item and PII scanning keep sourcing clean.
Paid work in your field.
Remote and flexible, with the rate stated before you accept. Every job is reviewed by a second expert, and approved work is paid automatically.
- Eight or more years of professional software engineering, or equivalent depth
- Staff, principal or architect-level design responsibility
- Experience operating production systems at meaningful scale
- Deep knowledge of specific languages, platforms or infrastructure
- Specialisms such as distributed systems, security, data or mobile
Write design documents
Design systems for realistic requirements and explain the trade-offs and rejected options.
Review code
Review pull requests for correctness, security and maintainability, as you would for your own team.
Grade agent trajectories
Step through coding agent runs and mark where and why the plan went wrong.
Review another engineer's work
Act as second reviewer on a colleague's designs and grades before delivery.
Software architecture: asked often.
What software architecture data can you deliver?
System design documents, architecture decision records, code reviews, agent trajectory grades, preference judgments and held-out design evaluation sets. Requests are scoped by language, platform, domain and difficulty, then matched to senior engineers verified in that area. A pilot batch lets your team check the bar before the request grows.
How do you verify senior engineers?
Every expert passes identity verification and a review of their experience and role history before receiving work, and may complete a short, paid design or review assessment. After that, every delivered item is reviewed by a second senior engineer, and label provenance records who created each judgment and who confirmed it.
Can I do this alongside a full-time engineering job?
Many experts do. There is no minimum commitment, work is remote and each job has a stated scope and deadline before you accept. You are responsible for checking your employer's outside-work and confidentiality policies, and you should never use your employer's code or confidential information in any job.
How do I get paid?
Rates are set by field and task and are shown before you accept a job. When a second expert approves your work, it is credited to your balance and paid automatically through Stripe once your identity and tax details are verified. Every job appears in an itemized ledger.
Where does the code come from?
Code is written new by engineers for the request or sourced from consenting creators who hold the rights, with consent bound to each item. Everything is fingerprinted at intake, checked for duplicates and scanned for personal information, and each item carries signed provenance so you can trace it.
Can we license the data exclusively?
Yes. Commissioned designs, reviews and evaluation sets can be licensed exclusively, with scope, term and exclusivity agreed in writing before work begins. Standing orders keep new items arriving at the same standard, which helps keep evaluation sets fresh as models improve and older items lose their value as a test.
Software architecture, done by experts.
Need expert-made data in this field, or have the expertise to make it? Start here.