Building PoC Agents in Minutes

#opensource #ai #softwareengineering

Last time I attended a manufacturing conference, I noticed that customers were eager for AI innovation in their businesses(They hungry to join the AI Race). They wanted customizable AI solutions that could be implemented immediately in their production chains. That’s why building quick prototypes or Proof of Concepts (PoCs) is exactly what AI engineers(Me) and software developers(Also me) need to focus on right now to my boss what we can do :))

For me, using a framework that Inherited multiple technologies is the key solution. When creating rapid prototypes for AI chatbots or AI agents, I use Parlant—a production-ready engine for AI chat agents that generates aligned responses and follows instructions.

Follow along with this example, and I’ll show you how to build an AI agent in just minutes using Parlant. (As long as you know Python, you’ll be good to go). I know an AI Engineer, it's me, just import Python library and it just run :D).

Also you need an AI Provider API Key (You can register on OpenAI Platform or Anthropic). For this example i will use my own Claude api key. (Put it in .env file)

For context, we will building a simple Travel Concierge name Danh Hoang Hieu Nghi (It's me, i love travel but no money)

if load_dotenv is not None:
    load_dotenv()

anthropic_key: Optional[str] = os.getenv("ANTHROPIC_API_KEY")
nlp_service = (
    p.NLPServices.anthropic if anthropic_key else p.NLPServices.openai
)

async with p.Server(nlp_service=nlp_service) as server:
    agent = await create_travel_concierge_agent(server)

Here the core heart of the Agent, we will create Agent, the condition and the action that the AI will follow (Guideline):

agent = await server.create_agent(
    name="Aurora Concierge",
    description="You are a friendly, detail-oriented travel concierge..."
)

await agent.create_guideline(
    condition="the customer greets you",
    action="Greet warmly... ask for destination, dates, budget, preferences."
)

To handle the whole flow without some infinite loop we will need a fall-back.

agent = await server.create_agent(
    name="Aurora Concierge",
    description="You are a friendly, detail-oriented travel concierge..."
)

await agent.create_guideline(
    condition="the customer greets you",
    action="Greet warmly... ask for destination, dates, budget, preferences."
)

You can customize your own coding my just adding new guideline :

await agent.create_guideline(
    condition="the customer asks for visa requirements",
    action="Summarize typical visa requirements and suggest official sources."
)