Frequently asked questions

Cadra is an AI capability assessment platform run by three agents on a learning engine: one writes a role-specific real-work task, one reads the working trail for authenticity and insight, and one scores the work with every score cited. People complete the task using AI tools, and organisations use the resulting evidence for hiring, training, and placement. A fourth agent that watches live work is planned but not part of the current pilot.

An AI capability assessment measures how effectively someone works with AI on a real task: what they delegate, how they prompt, what they verify, and the decisions they make along the way, rather than testing recall about AI tools. Cadra runs this as one realistic work brief completed with AI, with the full working session recorded and scored.

By reading the working process, not just the final output. Cadra scores multiple dimensions, for example technical depth, AI tool mastery, and engineering rigour, based on how the person framed the problem, prompted and verified the AI, and made decisions along the way. Every score has the supporting evidence attached.

Not in their traditional form. Standard coding tests are now solvable in minutes with AI, so passing proves tool access rather than judgment. Cadra takes the opposite approach: it expects people to use AI and scores how they use it: the full working session is recorded, and the scoring reads that process, not just the final output.

Cadra serves two groups today: hiring teams filling AI-native roles, and universities preparing cohorts for placement. A version for assessing and developing existing teams is coming soon.

The person receives a realistic work brief and completes it on their own machine, in their own IDE, typically within 90 minutes. While they work, Cadra captures their prompts and AI turns, design decisions, commits and diffs, and how they verified the result before calling it done.

No. Cadra uses no webcam, no lockdown browser, and no screen recording. Integrity comes from the work itself: the session records every prompt, edit, and decision, and no two candidates get the identical problem, so a coherent trail is hard to fake and easy to check.

Yes: using AI is the point. Candidates work on their own machine, in their own IDE, with whichever AI tools they normally use. The task is designed to be done with AI assistance, the way the real job is, and what gets scored is the judgment in how AI is used: what to delegate, what to verify, and when not to trust the output.

A report contains an overall fit score against the bar you set, a multi-dimensional breakdown with evidence cited behind every score, a behavioral profile of how the person worked, auto-generated first-round interview probes targeting weak or ambiguous areas, and a full audit trail where every score traces to a moment in the session.

By making AI use the thing being scored instead of the thing being policed. Tasks vary between candidates, scoring reads the working process rather than only the final answer, and every score traces to a specific moment in the recorded session, so a pasted-in solution without a coherent working trail scores poorly rather than well, with no webcam or lockdown browser needed.

Four days from setup to shortlist. Day 0 is a single 15-minute call to agree the rubric and pass mark; Cadra builds the role-specific task on day 1; candidates complete it at application stage on day 2; scored reports with a recommend/reject flag arrive by day 4.

Cadra currently runs role-specific assessments for AI Engineer, Forward Deployed Engineer, Data Scientist, ML Engineer, and Data Engineer. Each role gets its own task and rubric modelled on the real job, and new roles can be configured for a pilot.

One 15-minute call. You agree the rubric, the pass mark, and how much of the job the task covers; Cadra builds and maintains the task, runs candidates without scheduling or proctoring, and delivers scored reports. Configure a role once and every future cohort runs on the same rubric.

At application stage, before the first interview. Candidates complete the task without scheduling, and your first interview starts from a scored report with evidence and suggested probes, so senior engineers spend interview time on validated candidates instead of screening.

For AI-native roles, yes: Cadra replaces the screening stage. Instead of a puzzle that AI can solve for the candidate, applicants complete a role-specific work task with AI, and you receive scored reports with a recommend/reject flag before anyone books an interview. Rubric and pass mark stay under your control from a single 15-minute setup call.

With a baseline, a roadmap, and a re-baseline, not a one-off test. Cadra runs across a semester: students are baselined on industry-modelled tasks in week one, each gets a personal roadmap of what to fix in what order, and a re-baseline at the end of term shows measurable movement: evidence both the placement office and employers can use.

No. Cadra scores two tracks separately, because building AI systems and working with AI are different jobs. Track one assesses CS and engineering students on building and shipping AI systems; track two assesses business, analytics, and ops-track students on whether they are measurably better at the work because of AI.

Students get a capability profile backed by real work, a clear roadmap with runway to act on it before placement season, and interview-ready evidence. The institution gets a cohort benchmark by track, a signal on which gaps are systemic enough to teach, and term-over-term movement it can show.

Book a short call through the site at cadra.info/demo, or email hello@cadra.info. It takes about thirty minutes: we run a scenario relevant to you and tell you honestly whether Cadra fits your case.

Cadra is in early pilot, and pricing is scoped per engagement based on role count or cohort size. The intro call covers what a pilot for your case would involve and cost. There is no charge for the call or the scoping.