📌 Executive Concept Summary
This guide details the core technological pillars powering IntelliAssess AI:
- Form Generator Engine (Prompt-based & File-based generation).
- Question Libraries & My Uploads (Retrieval-Augmented Generation / RAG).
- AI Rubrics & Submission Checker (Automated scoring & manual overrides).
1. 🪄 AI Form Generator Engine & Model Selector
The Form Generator allows instructors and test managers to create complete assessments from prompt instructions or reference files.
A. Generation Modes
| Mode | Input Mechanism | Best Use Case |
|---|---|---|
| Prompt-Based | Topic text prompt + Magic Prompt enhancement | Quick test creation on general subjects or broad skill sets. |
| File-Based | PDF/DOCX upload or My Uploads library link | Testing candidates on proprietary training materials, resumes, or course syllabi. |
| Manual-Only / Hybrid | Manual placeholder slots + CSV import | Creating standardized tests with exact custom question stems. |
B. AI Model Architecture & Selection Matrix
IntelliAssess AI integrates with multiple AI providers (OpenAI, Groq, and Gemini) so users can select models based on context window, speed, and accuracy requirements:
| AI Model ID | Provider | Context Window | Generation Speed | Accuracy Level | Ideal Use Case |
|---|---|---|---|---|---|
gpt-5.4 | OpenAI | 1,000,000 tokens | Slower | Excellent | High-stakes technical assessments & complex reasoning essays. |
gpt-5-mini | OpenAI | 400,000 tokens | Fast | Very Good | Recommended default for form generation & bulk evaluation. |
gpt-4o-mini | OpenAI | 128,000 tokens | Fast | Very Good | Cost-effective, high-quality standard quiz generation. |
llama-3.3-70b-versatile | Groq | 128,000 tokens | Ultra-Fast | Very Good | Ultra-fast bulk generation for large question banks. |
gemini-3.1-pro-preview | Gemini | 1,000,000 tokens | Moderate | Excellent | Multimodal assessment generation across text, images, and audio. |
gemini-2.0-flash | Gemini | 1,000,000 tokens | Fast | Good | Default lightweight model for fast form & rubric generation. |
C. Supported Question Types & Data Schemas
IntelliAssess AI handles six distinct question formats:
┌──► Multiple Choice (mcq) ├──► Multi-Select (multiselect)IntelliAssess AI ├──► True / False (tf)Question Engine ├──► Short Answer (short_answer) ├──► Fill in the Blanks (fitb) └──► Descriptive / Essay (descriptive)
- Multiple Choice (
mcq): Single correct choice with randomized distractor options. - Multi-Select (
multiselect): Checkbox options requiring candidates to select all correct statements. - True / False (
tf): Binary boolean evaluation. - Short Answer (
short_answer): Single-line or brief response matched against model answer keys. - Fill in the Blanks (
fitb): Sentence stems containing inline blank inputs ([blank1],[blank2]). - Descriptive (
descriptive): Multi-paragraph open-ended responses auto-evaluated against multi-dimensional rubrics.
2. 📚 Question Libraries & My Uploads (RAG Indexing)
To prevent AI hallucination, IntelliAssess AI uses Retrieval-Augmented Generation (RAG) grounded in your uploaded reference materials.
Document Upload (PDF / DOCX) │ ▼Background Processing (Celery / Redis Worker) │ ▼Text Extraction & Semantic Chunking │ ▼Vector Embeddings (`pgvector` in PostgreSQL) │ ▼Context-Grounded Question Synthesis
RAG Pipeline Features:
- Background Ingestion Queue: Uploaded files in My Uploads are parsed asynchronously.
- Vector Embeddings (
pgvector): Document text is split into semantic chunks and converted into high-dimensional vector embeddings in PostgreSQL. - Semantic Context Retrieval: During question generation, the AI retrieves the top matching document passages, ensuring all question stems, distractors, and answer keys strictly reflect your uploaded source files.
3. ✍️ Per-Question AI Review Chat (“Generate with Feedback”)
The Generate with Feedback drawer allows users to refine individual questions interactively:
- Individual Question Threads: Open a dedicated AI conversation panel for any question card.
- Refinement Chips: One-click prompt suggestions such as “Make distractors harder”, “Add code formatting”, or “Simplify stem”.
- Safe Merge: Preview AI modifications and merge updates directly into your form without affecting other questions.
4. 📊 AI Rubrics & Submission Evaluation Engine
Automated evaluation in IntelliAssess AI balances AI speed with instructor oversight.
A. Multi-Criteria Rubric Anatomy
When evaluating Descriptive / Essay responses, the AI uses a structured rubric composed of:
- Criteria Tiers: Specific performance dimensions (e.g., Technical Accuracy, Clarity of Argument, Problem Solving).
- Point Distributions: Configurable weight percentages per criterion.
- Supporting Quotes & Evidence: The AI extracts exact quotes from candidate answers to explain why points were awarded or deducted.
B. Submission Checker & Grade Overrides
Candidate Submission │ ├──► Objective Questions (MCQ, FITB, T/F) ──► Instant Key Matching │ └──► Descriptive Questions ───────────────► AI Rubric Evaluation │ ▼ Instructor Override Review │ ▼ Final Candidate Scorecard
- Instant Objective Grading: MCQs, Multi-select, True/False, and Fill-in-the-blanks are scored instantly upon submission.
- Rubric-Based Essay Grading: Short answers and essays receive criterion scores and inline feedback notes.
- Manual Overrides: Instructors can override any score item, add manual comments, and recalculate total assessment scores in real time.
🏁 Summary Next Steps
Now that you understand the core concepts, proceed to the step-by-step tutorial: Quick Start: Creating & Inviting Users to Your First Assessment.
