The Evaluation Layer is the foundation of CrazyGoldFish’s AI reasoning stack. It automates the evaluation of handwritten, digital, audio, video, and diagram-based responses — ensuring fair, fast, and scalable assessment.
Why Evaluation Matters
- Problem: Manual grading is slow, inconsistent, and resource-intensive.
- Solution: The Evaluation Layer uses AI to evaluate multi modal submissions, deliver transparent, auditable results, and feed downstream remediation.
- Ecosystem Fit: It is the entry point of the loop — providing structured outputs that power the Personalization Layer (Action Plans) and Content Generation Layer (AI Studio).
Core Capabilities
- Multi modal Support → Handwritten answer sheets, typed responses, diagrams, audio (assignments).
- Dual Evaluation Modes → Rubric-based + Model-answer grading (used separately or combined).
- Auditable Workflows → Query handling, re-checks, annotated answer copies, compliance-ready exports.
- Scalability → Handles large batches and long responses with consistent accuracy.
- Dashboards → Track turnaround time (TAT), accuracy, and overall evaluation quality.
Components
📝 Exam Evaluation Suite
Automates grading of subjective exams with rubric + model-answer evaluation.
📘 Assignment Evaluation Suite
Handles daily classroom work, multi modal assignments (text, audio, video), and instant feedback.
Stakeholder Value
- Students → Receive annotated, transparent feedback in days → hours.
- Teachers → Save 7–10 hours/week on grading, focus more on teaching.
- Leaders → Get compliance-ready dashboards with turnaround + accuracy insights.
- Parents → Gain visibility into evaluation quality via annotated reports.
Ecosystem Integration
- Upstream: Exam/assignment submissions (handwritten, digital, audio, video).
- Core: Automated, rubric/model-answer evaluation with feedback JSON + annotations.
- Downstream: Outputs feed into Action Plan APIs (Student, Teacher, Parent), which then power AI Studio content generation.
Next Steps
- Dive into the Exam Evaluation Suite.
- Explore the Assignment Evaluation Suite.
FAQ
How does the Evaluation Layer automate grading across handwritten, digital, audio, video, and diagram-based submissions?
How does the Evaluation Layer automate grading across handwritten, digital, audio, video, and diagram-based submissions?
The Evaluation Layer ingests multimodal responses (text, handwriting, diagrams, audio, video), applies rubric and model‑answer logic, and returns scored outputs with clear feedback. A human‑in‑the‑loop review path handles edge cases for fairness and auditability, keeping educators in control while scaling assessment.
What does the Final Results API return—do you include section summaries, step‑wise marking, model answers, and an isApproved field?
What does the Final Results API return—do you include section summaries, step‑wise marking, model answers, and an isApproved field?
Yes. The Final Results API retrieves exam‑level scores, section summaries, step‑wise marking, and linked model answers, along with teacher feedback fields such as isApproved. These structured JSON outputs make it straightforward to embed transparent results in your product.
How do evaluation outputs flow into Personalization and AI Studio in the Evaluate → Personalize → Generate → Observe → Engage loop?
How do evaluation outputs flow into Personalization and AI Studio in the Evaluate → Personalize → Generate → Observe → Engage loop?
The Evaluation Layer is the first step in CrazyGoldFish’s closed‑loop AI Reasoning Layer. Its results feed downstream into Personalization for action plans and into AI Studio for standards‑aligned content generation, then into observation and engagement so insights continuously improve outcomes.
Can we deploy the Evaluation Layer as a white‑label layer inside our LMS or app while keeping our UI and brand?
Can we deploy the Evaluation Layer as a white‑label layer inside our LMS or app while keeping our UI and brand?
Yes—partners embed via the Embeddable UI or REST APIs while retaining their own UI and branding. You can push results to your LMS/ERP, trigger webhooks, and publish branded PDFs/CSVs so the experience stays native to your platform.
What accuracy should we expect and how is human‑in‑the‑loop review enforced for edge cases?
What accuracy should we expect and how is human‑in‑the‑loop review enforced for edge cases?
Teams typically target up to 95% accuracy when evaluations are aligned to rubrics and model answers. Human‑in‑the‑loop review, rechecks, and audit logs ensure teachers can validate or override results, preserving trust for subjective and high‑stakes scenarios.
What artifacts can we export or push to downstream systems after an evaluation run?
What artifacts can we export or push to downstream systems after an evaluation run?
You can export feedback JSON, annotated copies, and structured reports (e.g., PDFs/CSVs), and push updates to LMS/ERPs via webhooks or direct exports. These artifacts keep grading explainable and portable across your data workflows.
How are CBSE/ICSE curriculum alignment and GDPR handled in the Evaluation Layer workflows?
How are CBSE/ICSE curriculum alignment and GDPR handled in the Evaluation Layer workflows?
Workflows, rubrics, and reporting are designed to be CBSE/ICSE/GDPR aligned, with role‑based access and audit‑ready logs for compliance. This supports board‑specific expectations and privacy practices while keeping teachers in the approval loop.
What’s the recommended integration path—Embeddable UI versus REST APIs—for getting started with the Evaluation Layer?
What’s the recommended integration path—Embeddable UI versus REST APIs—for getting started with the Evaluation Layer?
Choose the white‑label Embeddable UI for a fast, low‑lift launch, or use REST APIs for deeper control over evaluations and exports. Both paths are API‑first and integrate with webhooks and LMS/ERP updates, so you can scale without specialized AI teams.
How do rechecks, appeals, and moderation work in the human‑in‑the‑loop flow?
How do rechecks, appeals, and moderation work in the human‑in‑the‑loop flow?
Rechecks can be initiated by staff or students, with rationale and outcomes captured in audit logs for transparency. Educators review AI‑proposed, rubric‑aligned scores and can approve, edit, or override before publishing, supporting defensible grading and moderation.
What teacher‑facing evidence and feedback do you return to keep grading explainable?
What teacher‑facing evidence and feedback do you return to keep grading explainable?
The system returns detailed feedback JSON, annotated copies, section‑level summaries, and step‑wise marking aligned to model answers and rubrics. These artifacts make strengths, gaps, and improvement points explicit for teachers and learners.