What the work involves
You will author migration tasks modelled on real production systems and then judge whether AI-generated migrations actually hold up. In practice that means reading unfamiliar legacy code, mapping dependencies, build systems, database schemas and integration points, and deciding what a safe migration path looks like — then writing the tests and validation strategy that would catch a behavioural difference rather than just a compile error. A large share of the time goes to evaluation: given a model's diff, you determine whether it preserves functionality, backwards compatibility, data integrity, security, and performance, and whether a competent reviewer would let it near production.
The distinguishing skill is not knowing that Java 8 became Java 17, but knowing where such a migration bites: default methods and interface resolution, removed internal APIs, date/time semantics, garbage collector behaviour under load, serialisation compatibility across a rolling deploy. The same applies across the other tracks — Python 2 to 3 text/bytes boundaries, Oracle-to-Postgres transaction and null-sorting semantics, SOAP-to-REST error contract loss, Hibernate lazy-loading behaviour that changes when you swap the ORM.
What the platform screens for
Mercor's screen is AI-led and follow-up heavy. It probes for migrations you personally executed — scale, timeline, what broke, how it was detected, how it was rolled back — and it will keep drilling on any answer that stays at the architecture-diagram level. Expect scenario questions where you are handed a plausible-looking migration diff and asked what you would reject and why, plus questions about staged rollout, dual-write and shadow-read patterns, and how you establish behavioural equivalence when the legacy system has no meaningful test suite.
- Depth in at least one migration track, verifiable to specifics
- Evidence of ownership on production systems where regression carried real cost
- Written reasoning that a strong engineer would accept as review commentary
- Regulated or high-reliability domains — banking, payments, trading, insurance, healthcare, telecom — are weighted especially heavily, as is mainframe experience with z/OS, CICS, JCL, DB2, and VSAM
Logistics
Fully remote and largely asynchronous, with work delivered against task batches rather than fixed shifts. Most contributors treat this as part-time alongside a primary role; sustained availability of roughly 10–20 hours a week is typical, and volume fluctuates with the research program's needs. Rates around $200/hr have been observed for this listing; final terms are set by the platform per contributor and are not guaranteed. No open-source maintainer track record is required — enterprise migration experience counts fully.