Learn to Generate Synthetic Data with CAMEL Multi-Agent Role-Playing

How to Generate Synthetic Data Using CAMEL: A Multi-Agent Role-Playing Approach

You need 2,000 training dialogues for a corporate assistant? Writing them manually would take weeks. Or you want AI agents to discuss service architecture and identify bottlenecks. The CAMEL AI framework (Communicative Agents for 'Mind' Exploration of Large Scale Language Model Society) solves these through role-based dialogues between LLM agents. Our engineers hold AWS certification and have 10+ years of multi-agent system experience — we've deployed CAMEL in over 25 successful projects, saving clients up to $20,000 per project. We offer a 100% satisfaction guarantee and our process adheres to ISO 27001 standards. Custom agent development starts at $5,000, with full pipeline solutions averaging $25,000. Contact us for a consultation.

What Problems Does Role-Playing Solve and How to Reduce Hallucinations?

Lack of dialogue data for fine-tuning: many companies have only documents, not labeled dialogues. CAMEL generates structured Q&A conversations from your materials, providing high-quality fine-tuning data. Hallucinations and incomplete answers: agents clarify each other, reducing errors. In CAMEL we set up a critical cycle: one agent checks another's answer, points out inconsistencies, and asks for reformulation. This cuts errors by 35% in our tests. We also use retry logic with a limit on attempts and response format validation. Multi-step reasoning: role-play forces agents to work step by step, improving reasoning quality.

Our Implementation: A Case Study

In our practice, a client wanted to fine-tune a corporate assistant on tech support documentation. No manual dialogues — only 80 PDF guides. We configured Role-Playing: role "New Employee" asks questions, "Experienced Engineer" answers using the document text. In 3 days we generated 3,200 dialogues; fine-tuning on GPT-4o-mini improved answer quality by 28%.

What's Included and Technology Stack

• Audit of your data and use cases.
• Role and prompt design tailored to your specifics.
• Development of custom agent tools and capabilities (search, test execution).
• Synthetic data generation in required format (ShareGPT, Alpaca, OpenAI).
• Agent integration into your pipeline and team training.

We use Python 3.12+ with the camel-ai framework. Models: GPT-4o (default), Claude 3.5, LLaMA 3. For embeddings: OpenAI Embeddings (1536-dim) or Hugging Face. Version control via Git, experiments logged in MLflow.

# Basic Role-Playing (key settings)
from camel.agents import RolePlaying
from camel.types import ModelType, TaskType

role_play_session = RolePlaying(
    assistant_role_name="Python developer",
    user_role_name="Product Manager",
    assistant_agent_kwargs={"model": ModelType.GPT_4O},
    user_agent_kwargs={"model": ModelType.GPT_4O},
    task_prompt="""Develop an implementation plan for a notification service API.
Requirements: WebSocket for real-time, REST for CRUD, support for push/email/SMS channels.""",
    with_task_specify=True,
    task_type=TaskType.AI_SOCIETY,
)

init_assistant_msg, init_user_msg = role_play_session.init_chat()

# Custom Agents with Tools
from camel.agents import ChatAgent
from camel.messages import BaseMessage
from camel.types import ModelType
from camel.toolkits import OpenAIFunction

def search_codebase(query: str, file_pattern: str = "*.py") -> str:
    """Search the project codebase."""
    results = code_search.search(query=query, pattern=file_pattern)
    return str(results[:5])

def run_tests(test_file: str) -> str:
    """Run tests and return result."""
    result = subprocess.run(["pytest", test_file, "-v"], capture_output=True, text=True)
    return result.stdout[-2000:]

system_message = BaseMessage.make_assistant_message(
    role_name="Senior Software Engineer",
    content="""You are an experienced software engineer.
Use tools to analyze code and write solutions.""",
)

engineer_agent = ChatAgent(
    system_message=system_message,
    model=ModelType.GPT_4O,
    tools=[
        OpenAIFunction(search_codebase),
        OpenAIFunction(run_tests),
    ],
)

Why CAMEL and Typical Agent Roles

CAMEL outperforms other frameworks (e.g., AutoGen or CrewAI) in tasks requiring role asymmetry and deterministic dialogue format. It provides built-in dialogue termination validation and flexible message count configuration. In our tests, CAMEL generates synthetic data 40% faster than AutoGen with equal quality.

Process, Common Problems, and Timeline

1. Analysis — understand your task, define roles and agent goals.
2. Design — write prompts, configure tools.
3. Implementation — code in Python with camel-ai, test on a small set.
4. Testing — check dialogue quality, answer accuracy, execution time.
5. Deployment — deploy agents in your infrastructure (Kubernetes, SageMaker).

Our role-playing AI agents can simulate realistic conversations. We also support autonomous agents that work without human intervention and specialize in building GPT-4o agents for high-quality dialogues. Automated training data generation saves time and costs. Contact us — we'll analyze your scenario and provide a solution. Request a consultation today.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.