AI-Powered Automatic Test Generation System

Instructors spend up to 40% of their time creating tests, while students easily find ready-made answers. Manually composing a 30-question exam takes 4–6 hours—and that's just one variant. Developing an AI system for automatic test generation pays off within months: instructor time savings of up to 70%, and test cost reduced to 10–20% of manual labor. Typical project cost ranges from $5,000 to $15,000, with annual savings of $20,000 or more from reduced instructor workload. We offer a solution: a system that generates unique exam tasks based on your material, considering difficulty levels and Bloom's taxonomy. Each student gets their own variant, and grading of open-ended answers is automated. The system integrates with Moodle, iSpring, and supports custom export formats. Get a consultation on your project—we will evaluate automation opportunities in one day.

Why Traditional Tests No Longer Work

Manual question creation consumes hours, and question banks quickly become outdated. Students share answers, and instructors spend time grading. AI generation solves both problems: it creates an infinite number of variants and evaluates open-ended answers in seconds. According to a McKinsey study, automating routine tasks frees up to 30% of instructor time.

How AI Generates Questions by Bloom's Taxonomy

We use GPT-4o with prompts tailored to each level of Bloom's taxonomy: from recalling facts to creating new products. The code below shows the implementation.

from openai import AsyncOpenAI
from enum import Enum
import json

client = AsyncOpenAI()

class BloomLevel(Enum):
    REMEMBER = "remember"
    UNDERSTAND = "understand"
    APPLY = "apply"
    ANALYZE = "analyze"
    EVALUATE = "evaluate"
    CREATE = "create"

BLOOM_PROMPTS = {
    BloomLevel.REMEMBER: "Create a question on recalling facts, dates, definitions",
    BloomLevel.UNDERSTAND: "Create a question on understanding: explanation, paraphrasing, examples",
    BloomLevel.APPLY: "Create a practical task: applying knowledge in a new situation",
    BloomLevel.ANALYZE: "Create a question on analysis: comparison, identifying causes, structuring",
    BloomLevel.EVALUATE: "Create a question on evaluation: justifying a judgment, critiquing an approach",
    BloomLevel.CREATE: "Create a task on synthesis: developing a solution, creating a product",
}

async def generate_question(
    topic: str,
    source_text: str,
    question_type: str,
    bloom_level: BloomLevel = BloomLevel.UNDERSTAND,
    difficulty: str = "medium"
) -> dict:
    response = await client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "system",
            "content": f"""Create a test question.
            Type: {question_type}.
            Bloom's taxonomy level: {bloom_level.value}. {BLOOM_PROMPTS[bloom_level]}.
            Difficulty: {difficulty}.

            For multiple_choice: 4 options, 1 correct, 3 plausible distractors.
            For open_answer: model answer + scoring rubric.
            For case_study: scenario + 3-5 questions at different levels.

            Return JSON: {{
                question: "question text",
                type: "{question_type}",
                bloom_level: "{bloom_level.value}",
                options: ["A...", "B...", ...],
                correct_answer: "...",
                explanation: "why this answer",
                scoring_rubric: {{...}}
            }}"""
        }, {
            "role": "user",
            "content": f"Topic: {topic}\n\nMaterial:\n{source_text[:2000]}"
        }],
        response_format={"type": "json_object"}
    )
    return json.loads(response.choices[0].message.content)

Generating a Complete Exam Variant

We assemble a test from questions with a specified distribution across levels. For example, 30% understanding, 30% application, 20% analysis, and 20% recall.

async def generate_exam_variant(
    course_topics: list[str],
    total_questions: int = 30,
    time_limit_min: int = 60,
    bloom_distribution: dict = None
) -> dict:
    if not bloom_distribution:
        bloom_distribution = {
            BloomLevel.REMEMBER: 0.2,
            BloomLevel.UNDERSTAND: 0.3,
            BloomLevel.APPLY: 0.3,
            BloomLevel.ANALYZE: 0.2
        }

    questions_by_level = {
        level: int(total_questions * fraction)
        for level, fraction in bloom_distribution.items()
    }

    all_questions = []
    tasks = []

    for level, count in questions_by_level.items():
        for i in range(count):
            topic = course_topics[i % len(course_topics)]
            q_type = "multiple_choice" if level in [BloomLevel.REMEMBER, BloomLevel.UNDERSTAND] else "open_answer"
            tasks.append(generate_question(
                topic=topic,
                source_text="",
                question_type=q_type,
                bloom_level=level
            ))

    all_questions = await asyncio.gather(*tasks)

    return {
        "variant_id": f"V{random.randint(1000, 9999)}",
        "time_limit_min": time_limit_min,
        "total_points": sum(q.get("points", 1) for q in all_questions),
        "questions": list(all_questions),
        "bloom_distribution": {l.value: c for l, c in questions_by_level.items()}
    }

Comparison: AI vs. Manual Method

AI generation is 10x faster: 30 questions in 2 minutes instead of 4–6 hours. Test generation cost is 5–10x lower, and the number of unique variants reaches up to 30 per question. AI-powered grading is 100x faster than manual grading.

How to Ensure Variant Uniqueness?

Each question is not a copy. The system generates up to 30 unique versions by varying numbers, names, context, and answer order. Difficulty remains consistent, verified across dozens of projects.

async def generate_unique_variants(
    base_question: str,
    n_variants: int = 30,
    maintain_difficulty: bool = True
) -> list[dict]:
    response = await client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "system",
            "content": f"""Create {n_variants} unique versions of the question.
            Vary: numbers, names, context, order of answer options.
            Difficulty {'must remain the same' if maintain_difficulty else 'may vary'}.
            Return a JSON array."""
        }, {
            "role": "user",
            "content": f"Original question: {base_question}"
        }],
        response_format={"type": "json_object"}
    )
    return json.loads(response.choices[0].message.content)["variants"]

Automatic Grading of Open-Ended Answers

For open-ended questions, an AI grader compares the student's answer against the rubric and model answer, assigning points with comments.

async def auto_grade_open_answer(
    question: str,
    correct_answer: str,
    rubric: dict,
    student_answer: str
) -> dict:
    response = await client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "system",
            "content": f"""Grade the student's answer according to the rubric.
            Question: {question}
            Model answer: {correct_answer}
            Scoring criteria: {json.dumps(rubric, ensure_ascii=False)}

            Evaluate the answer and return JSON:
            {{score: 0-100, feedback: "detailed feedback", strengths: [], weaknesses: []}}"""
        }, {
            "role": "user",
            "content": f"Student answer: {student_answer}"
        }],
        response_format={"type": "json_object"}
    )
    return json.loads(response.choices[0].message.content)

Ensemble-Based Evaluation

For reliable grading of disputed answers, we use an ensemble of three different AI models. If two models give the same score, it is final. In case of disagreement, a third model is consulted, and the answer is flagged for manual review. This reduces the error rate to less than 1% per test.

What Is Included in the Work

• Designing the generation architecture for your content
• Developing prompts for all Bloom's taxonomy levels
• Integration with LMS (Moodle, iSpring, custom API)
• Generation of up to 30 unique variants per question
• Automatic grading module for open-ended answers with rubrics
• Documentation
• Instructor training
• 2 months of support
• Access rights to the system

Process of Work

1. Analysis — we study your educational material and test requirements
2. Design — determine question types, distribution across levels, export format
3. Implementation — write generation code, integrate with your LMS
4. Testing — validate on real students, adjust prompts
5. Deployment — launch into production, hand over documentation

Indicative Timelines

• Test generator from ready textual material: 1–2 weeks
• Full platform with auto-grading, analytics, and LMS integration: 2–3 months

The cost is calculated individually based on your data volume and required integrations. Budget savings on training amount to 30–50% of manual testing costs. Investment in automation pays back within 2 months. Evaluate your project in one day — order an individual demo to discuss details.

With over 5 years of experience in AI/ML and 20+ content generation projects, we guarantee stable operation and timely delivery. Get a consultation on your project — we will evaluate automation opportunities in one day.

Generative AI Development: From Prompt to Production API

We often receive a task "generate a product image" — on the surface it seems simple. But behind this lies a choice between dozens of models, configuring the inference pipeline, manually solving consistency issues, integrating into the product backend, and answering why the model generates hands with six fingers in staging but not in production. Let's break down the directions we work with.

Image Generation: From Prompt to Production API

The current landscape includes FLUX.1 [dev/schnell/pro] from Black Forest Labs and Stable Diffusion 3.5. FLUX.1 [schnell] takes 4 steps instead of 20–50 for SDXL — 5–12 times faster — while maintaining higher quality. On an A100 80GB — 1.2–1.8 s per 1024×1024 image at batch_size=4.

A typical deployment issue: FLUX.1 [dev] requires 24+ GB VRAM in fp16. On A10G 24GB it fits tightly; at batch_size>1 — OOM. Solution: torch_dtype=torch.bfloat16 + enable_model_cpu_offload() from diffusers, or quantization via bitsandbytes to NF4 — minimal quality drop, memory consumption drops to 12–14 GB.

ControlNet and IP-Adapter are key tools for production tasks where controllability is needed. ControlNet with Canny/Depth/Pose maps provides structural control. IP-Adapter (especially IP-Adapter-FaceID) allows transferring character identity to generations — this is the foundation for personalized content. More about ControlNet can be found on Wikipedia.

Case study: e-commerce photography. A retailer with 8000 SKUs needed lifestyle photos for each product. Pipeline: product segmentation (Segment Anything Model 2) → background removal → inpainting with FLUX.1 [dev] using product image as IP-Adapter reference → upscale via RealESRGAN_x4plus. The generation cost is negligible compared to professional photography, providing huge savings. Throughput — 200 images/hour on 2× A100. Our extensive experience from 30+ projects ensures we select the optimal model for your task — an evaluation can be obtained upfront.

Why Is Model Selection Only Half the Battle?

Fine-tuning for a Specific Style or Character

Dreambooth and LoRA are the standard for adapting to a specific visual style or object. LoRA trains in 2–4 hours on 20–30 reference images on a single A100. Rank 16–32 is usually sufficient for style; rank 64+ is needed for precise face reproduction.

A common mistake: training LoRA too long — the model overfits to references, losing the ability to vary. Sign: at cfg_scale=7, all images look like copy-paste of references. Solved by early stopping (usually 1500–2000 steps for 20 images) and prior_preservation_loss.

For deeper customization — full fine-tuning via diffusers + accelerate with FSDP on multiple GPUs. But that already takes 40–80 hours of training and requires a truly large dataset (1000+ images).

Comparison of Image Generation Approaches

Video Generation: Which Models Are Best?

Open-weight video models still lag behind commercial ones in stability and length. Wan2.1 is the best choice for self-hosting: 14B parameters, runs on 2× A100, delivers acceptable quality for short clips.

The main pain of video generation is temporal consistency: the character changes clothing color at the third second, objects "drift." Partial solution — generation with motion_bucket_id and noise_aug_strength in Stable Video Diffusion, or using I2V (image-to-video) instead of pure text-to-video. As noted in VideoPoet research, consistency is achieved by training on long sequences.

AnimateDiff remains a working tool for short loops and motion effects on top of SD/FLUX. Not Sora, but deployable locally and predictable.

Music and Audio Generation

AudioCraft from Meta (MusicGen + AudioGen) is a production-ready stack for music generation. musicgen-large (3.3B) generates 30 s of music in ~8 s on A100. Control via text prompt and melody conditioning — you can specify a melody by humming.

Stable Audio Open from Stability AI is an alternative with length up to 47 s, better structural control (intro/verse/chorus). Deployment is similar: diffusers + FastAPI.

For voice-over and dubbing — ElevenLabs API or self-hosted XTTS v2 (see Speech AI service). For sound design and foley — AudioGen.

3D Generation: Current Practical State

3D generation has not yet reached the same maturity as 2D. But for specific tasks, tools are already working:

TripoSG and Shap-E — text/image-to-3D. Shap-E from OpenAI generates simple 3D meshes in seconds, but geometry is rough. TripoSG gives more detailed results but requires post-processing (remeshing, UV unwrapping).

Wonder3D and Zero123++ — 3D reconstruction from a single image. They work by generating multi-views (6–8 views) and then 3D reconstruction via NeuS or instant-ngp.

Gaussian Splatting (3DGS) — not generation, but reconstruction from a series of photos/videos. For product cards and real estate it's already production: 50–200 photos → 3DGS model in 15–30 min on RTX 4090 → interactive 3D viewer in browser.

What Infrastructure Is Needed for Generative AI Deployment?

Critical for generative models:

• Task queue — Celery + Redis or Ray Serve. Synchronous HTTP for image generation is unacceptable with >5 concurrent requests.
• Caching — similar prompts yield similar results. Semantic cache via embeddings (faiss + sentence-transformers) can reduce GPU load by 20–40%.
• Quality monitoring — CLIP score for text-image alignment, FID for evaluating generation distribution. Integrate into MLflow or Weights & Biases.
• Storage — generated images immediately to S3/MinIO, not on the inference server disk.

What's Included in the Deliverables

We take the project turnkey — from model selection to deployment and monitoring. The result includes:

• Model (or API integration) with performance benchmarks (latency p99, throughput).
• Pipeline documentation (prompt engineering guide, model card, dependency versions).
• Integration with your backend (REST/gRPC, queues).
• Configured monitoring (dashboards, alerts for quality drift).
• Training workshop for the team (2–4 hours).
• Warranty support for 3 months after launch — as part of our quality certificate.

We have completed 30+ projects in generative AI — this gives us the right to guarantee results.

How Is the Generative AI Development Process Structured?

1. Analysis (1–2 days): audit of current architecture, clarification of use case, selection of models and success metrics. We evaluate the project free of charge.
2. Proof of Concept (1–3 weeks): quick prototype on your data — to see real quality, not blog demos.
3. Design (1–2 weeks): pipeline architecture, infrastructure (GPU cluster/API), A/B testing plan.
4. Implementation and fine-tuning (4–12 weeks): development, LoRA/full fine-tuning, integration with queue and cache.
5. Testing (1–2 weeks): load tests, metric validation, edge-case verification (negative scenarios).
6. Deployment and monitoring (1–2 weeks): production deployment, monitoring setup, documentation.

Timeline Estimates

Integration of a ready API (DALL·E 3, Midjourney API, Stability API) — 1–2 weeks. Self-hosted pipeline with fine-tuning — 6–12 weeks. Full platform with UI, queues and monitoring — 3–6 months. The specific cost is calculated individually after analyzing your scenario.

Contact us — order a consultation, and we will select the optimal architecture for your project. Get a preliminary cost and timeline estimate for free.