AIVIA’s evaluator is dynamic — it adapts to every candidate’s responses in real time. The Context Pack is what gives that dynamic conversation its focus. It tells the system what matters for this specific role and component, what to probe, and how to interpret the candidate’s responses.
01 Two Configuration Groups
Every Context Pack is built from two groups of settings:
Global Constraints
Shape difficulty, tone & scoring
Expected Background. Beginner, Intermediate, or Advanced — calibrates question difficulty and scoring thresholds.
Probe Style. The question patterns that control how the AI challenges candidates — “Walk me through…”, “What broke when…”, “Trade-off between…”, “Why not…”, “How would you teach…”, “Scale this up…” Each style extracts different kinds of signal.
AI Persona. The interviewer style that shapes depth, tone, and follow-ups. Options include Curious Senior Colleague (collaborative, asks “why”), Technical Detective (probing, seeks edge cases), Product-Minded Engineer (focuses on impact and trade-offs), and Academic Researcher (theory-first, deep rigor).
Component-Specific Controls
Derived from component failure modes
Skill Probes. The specific technical skills the evaluation should target — derived from the component’s real-world failure modes.
Scenario Cards. The situations the candidate will reason through — built from symptoms, artifacts, and false leads tied to the component.
Evidence Style. What counts as strong versus weak evidence in responses. The evaluator continuously checks answers against these requirements and triggers follow-up probes when evidence is missing or vague.
02 How It Prevents Drift
Without a strong configuration layer, automated evaluation drifts — questions become generic, scoring loses meaning, and the outcome stops reflecting the real work.
The Context Pack prevents this through a continuous evidence loop. The evaluator doesn’t just ask questions and move on. It checks every response against the defined evidence requirements, detects gaps or vague answers, and triggers targeted follow-up probes to extract the missing signal. This happens automatically, without human supervision.
The result: high-signal reports where every score is grounded in evidence the candidate actually provided.
03 Who Configures It
Every scenario post comes with a default Context Pack — smart defaults derived from the component and its failure modes. For most candidates taking evaluations from the open library, the defaults work as-is.
Hiring teams creating custom evaluations can edit Context Pack fields directly — adjusting expected background, probe style, AI persona, skill probes, and scenario focus to match their specific role. The hiring team’s configuration remains private; other teams and candidates cannot see it.
The Context Pack is patent pending. For a full interactive walkthrough of probe styles, AI personas, and scenario cards, see the Context Pack page.