AI Assistants

Harness artificial intelligence to automate audience engagement.

AI assistants, formerly known as chatbots, are the first line of support to triage, answer, and solve your audience's questions, issues, and concerns. All chats begin with an AI assistant.

---

Chatbot vs LLM

Scripted chatbots have long been a staple in customer service automation, preceding the modern large language models (LLMs) that have followed. LLMs introduced important advancements for audience engagement, like unstructured conversations and the voice and vision modalities.

Proto AICX marries the best parts of each technology:

Capability	Chatbots	LLMs	Proto AI Assistants
Content privacy	✔		✔
Scripted chats	✔		✔
Runs in messaging apps	✔		✔
Human handoffs	✔		✔
Trainable on resources		✔	✔
Unstructured chats		✔	✔
Voice chats		✔	✔
Realtime language translation		✔	✔

How Assistants Work

ProtoAI

The combination of features from traditional chatbots and modern LLMs is thanks to ProtoAI, our natural language processing (NLP) engine. It assesses all inbound messages at runtime, using sophisticated NLP pattern recognition to match the person's conversational intent with an appropriate action in the assistant's scripted flow.

If there is no match, and a specified LLM has been enabled for the assistant, ProtoAI will send the message through to the LLM for an informed, organic response.

By empowering your assistants with both a scripted flow and a trained, personalised LLM, you'll have an incredibly robust, well-rounded AI at your disposal.

---

Chat Flow

AI assistants are interacted with through realtime messaging, and can be configured to handoff chats to a human. For human-to-human emailing, see tickets.

Assistants run this sequence for every inbound text or voice message received:

1

Receive message

A message is received through one of the assistant's enabled messaging apps or website interfaces.

2

Check triggers, apply actions

ProtoAI checks for scripted triggers from a system event or which match the person's intent, and applies corresponding actions—like a stock reply or human handoff. The flow ends here if any trigger/action is applied.

3

Use Fallback trigger (if enabled)

The Fallback trigger is a catch-all for unmatched conversational intent. It will intercept the message and reply with a scripted response, ending the flow.

4

Use LLM (if enabled)

If nothing was triggered in steps 2 & 3, the message is sent to a specified LLM, which generates an unstructured response. LLMs can be trained and personlised for your use case.