Services

AI Agents & Chatbots in Adelaide

Custom AI assistants and agents for customer service, internal knowledge and process automation.

Discuss your projectBook a consultation or learning sessions

AI Agents & Chatbots for South Australian organisations

Customers expect answers instantly and around the clock. Staff lose hours each week hunting for policies, procedures and past answers. Modern AI agents and chatbots can handle routine questions, guide people through processes, and complete multi-step tasks, while handing over to a person when they should.

EITSAA builds AI chatbots and agents for Adelaide businesses and agencies that are grounded in your approved content, not the open internet. Every assistant includes guardrails, logging and human-in-the-loop controls, so you stay in control of what it says and does.

Our AI agents answer from your approved content, hand over to people when they should, and log every interaction for quality and compliance.

What's included

  • Customer service chatbots for websites, Teams and Slack
  • Internal knowledge assistants grounded in your documents
  • Agentic workflows that automate multi-step tasks with human approval
  • Voice and email triage bots
  • Conversation design with intents, slot filling, validation and retries
  • Guardrails, logging and human-in-the-loop design

Live demo

Meet our Mini Bot

A small working assistant that shows how a well-designed chatbot behaves in your industry. Choose local government, healthcare, insurance and professional finance, retail, professional services or education, and it runs that sector's conversation. It works out what a visitor wants, asks for the missing details one question at a time, checks every answer, retries with clearer wording and hands over to a person when it should. Open the trace panel to watch each decision as it happens.

  • Six industry flows
  • Intent detection
  • Slot filling
  • Validation and retry
  • LLM-assisted steps
  • Guardrails
  • Live trace

Flow control: how the conversation moves

Flow control covers intents, slots, branches, retries and routing. It decides what to ask next and when to hand over.

  1. Choose your industry

    Rules

    The visitor picks a sector, and can switch at any time. Each industry loads its own intents, questions, branches and guardrail profile.

    “Local government”

  2. Detect intent

    Rules

    Work out whether the visitor wants to book, ask a question, get pricing or reach a person.

    “I'd like a chatbot for our website”

  3. Elicit the slots

    Rules and LLM

    Ask for each missing detail, one clear question at a time. Answers can branch the flow, and details already given are never asked again.

    “Roughly how many people will use it?”

  4. Validate and retry

    Rules

    Check each answer. If it cannot be used, re-ask with more helpful wording. After three tries, hand over gracefully.

    “A rough number helps, like 25.”

  5. Call the LLM between questions

    LLM

    A language model rephrases the next question, reads free-text answers into slots and writes the summary. Every reply is moderated.

    “Lovely. And which day suits you?”

  6. Confirm and hand off

    Rules

    Read the answers back, let the visitor change any of them, then continue to the contact form. Safety rules hand over to a person at once.

    “Does that look right?”

A sample path through each industry flow

  • Local governmentstandard guardrails

    1. Report a problem
    2. Type of problem
    3. Location
    4. Safety risk?
    5. Yes: hazard details
    6. Confirm

    Records request, Complaint, Rates, Permit or approval, Service request

  • Healthcare and aged carestrict guardrails

    1. Book an appointment
    2. Type
    3. For whom?
    4. Someone else: consent
    5. Day and time
    6. Confirm

    Referral, New patient intake, Care enquiry, Appointment

  • Insurance and professional financestrict guardrails

    1. Lodge a claim
    2. Policyholder or client?
    3. Acting for someone: authority
    4. Type of claim
    5. Anyone injured? ?Yes: priority
    6. Confirm

    Payment arrangement, Invoice enquiry, Claim, Quote

  • Retail and e-commercestandard guardrails

    1. Return or exchange
    2. Order number
    3. Reason
    4. Faulty: photo?
    5. Refund or exchange
    6. Confirm

    Payment issue, Return or exchange, Order status, Product advice

  • Professional services and SMEsstandard guardrails

    1. Request a quote
    2. What you need
    3. Project size
    4. Budget set?
    5. Yes: budget range
    6. Confirm

    Meeting, Quote request, Enquiry

  • Education and trainingstandard guardrails

    1. Enrol
    2. Domestic or international?
    3. International: visa help
    4. Level
    5. Intake
    6. Confirm

    Student support, Enrolment, Course enquiry

Guardrails: the controls around the conversation

Flow control decides what to ask next. Guardrails decide what may come in, what the model may see and do, and what a visitor may be shown. Every decision is logged and a route to a person stays open at each step.

Visitor message

  1. 1Input

    8 controls

    • IN-01 Input limits
    • IN-02 Rate limits
    • IN-03 Toxicity and profanity filter
    • IN-04 Prompt-injection and jailbreak detection
    • IN-05 System-prompt extraction detection
    • IN-06 Sensitive-data detection and redaction
    • IN-07 Topic and scope enforcement
    • IN-08 Unsafe-advice refusal

  2. 2Context

    2 controls

    • CX-01 Allow-listed context
    • CX-02 Data minimisation

  3. 3Model

    5 controls

    • MD-01 Fixed server-side system prompt
    • MD-02 Token, temperature and time caps
    • MD-03 Confidence threshold
    • MD-04 Retry, fallback model and graceful degrade
    • MD-05 Evaluation set and pinned model

  4. 4Tool and action

    2 controls

    • TL-01 Action allow-list
    • TL-02 Least-privilege execution

  5. 5Output

    4 controls

    • OT-01 Output moderation
    • OT-02 Output schema validation
    • OT-03 System-prompt leakage check
    • OT-04 Grounding check

  6. 6Human oversight

    4 controls

    • OV-01 Vulnerable-customer and crisis escalation
    • OV-02 Retry limit and human hand-off
    • OV-03 Audit logging
    • OV-04 Disclosure and route to a person

OV-03 audit logging with masked values spans every layer, and OV-04 keeps a route to a person open at every step

Failure modes we design for

Failure modeWhat goes wrong
HallucinationA reply states something that was not supplied or is not true.
Prompt injection and jailbreakText tries to override instructions or change the bot's role.
System-prompt leakageThe bot reveals its hidden instructions.
Sensitive-data leakagePersonal, health or payment data is sent to a model, stored or repeated.

Each control is mapped to the Voluntary AI Safety Standard, the OWASP Top 10 for LLM Applications and the NIST AI Risk Management Framework, and tuned per industry. To see the full control set for your organisation, contact us, or read more in guardrails for customer-facing chatbots.

Know more: talk to us

How we can help

Website customer service bot

Answers common questions, captures leads and books appointments around the clock, with live hand-over to your team.

Internal knowledge assistant

Helps staff find answers in policies, procedures, contracts and SharePoint in seconds, with source citations.

Microsoft Teams and Slack agents

Bring AI into the tools your team already uses for HR, IT and operations questions.

Agentic process automation

Agents that read emails, extract data, update systems and draft responses for human approval.

Email and ticket triage

Automatically classify, route and prioritise incoming requests.

Our delivery process

  1. Discover

    We clarify goals, users, constraints and success measures, and review your current systems and data.

  2. Design

    We agree the architecture, security controls and user experience before the build begins.

  3. Deliver

    Fortnightly sprints with demos, automated testing and continuous deployment.

  4. Operate and improve

    Monitoring, support and a backlog of improvements based on real usage.

Technology we use

Microsoft Copilot Studio, Azure AI, AWS Bedrock Agents, Amazon Lex, LangGraph, vector databases, Teams, Slack, web widgets.

Why choose an Adelaide partner?

Working with a local team means face-to-face workshops, an understanding of South Australian regulatory and procurement requirements, and support in your time zone. Engaging EITSAA also keeps investment and jobs in South Australia, which counts under the SA Industry Participation Policy in government tenders.

See our government capability statement or read how we work.

Outcomes you can expect

Every AI Agents & Chatbots engagement starts with outcomes we agree on, not just a list of features. Before we begin, we set measures such as hours saved, faster turnaround, higher customer satisfaction, lower operating cost or reduced risk. We report against them throughout delivery. You get a fortnightly demonstration of working software, a transparent backlog, and a named Adelaide-based lead who is accountable for progress.

When the work is done, you keep the capability. Documentation, source code, infrastructure definitions and runbooks are handed over in full, and knowledge-transfer sessions give your team the confidence to run and extend what we have built.

Security, privacy and compliance

Security and privacy are designed in from the first workshop. Our AI Agents & Chatbots solutions follow the principle of least privilege, encrypt data in transit and at rest, and can be hosted in Australian cloud regions to meet data sovereignty requirements. We align with the ACSC Essential Eight, the Australian Privacy Principles and, for public sector clients, the SA Government Information Security Management Framework.

For regulated or sensitive work, we can complete privacy impact assessments, provide security documentation for your risk team, and support independent penetration testing before go-live.

Frequently asked questions

Can I try a chatbot before we talk?

Yes. The Mini Bot on this page is a working demo. It detects what you want, asks for missing details, validates your answers, retries with clearer wording and hands over to a person, and its trace panel shows each decision.

Will the chatbot make things up?

We reduce this risk by grounding answers in your approved documents, requiring citations, and testing against an evaluation set before launch. The bot is designed to say when it doesn't know.

How long does a chatbot take to build?

A focused proof of value typically takes four to six weeks, including content preparation, testing and a pilot with real users.

Can it hand over to a human?

Yes. Hand-over rules are part of every design, and conversations can pass to live chat, email or your ticketing system.

What does it cost to run?

Running costs depend on usage. We design for cost efficiency and provide monthly usage reporting.

Further reading

Plain-English guides from our Insights library related to AI Agents & Chatbots.

Browse all insights →

Learn

Reference architectures for AI Agents & Chatbots

Related services

AI Governance & Assurance

Governance, guardrails and assurance for every kind of AI: machine learning, generative AI, LLMs, agents and automated decisioning, aligned to Australian guidance and international standards.

Explore AI governance →

Training & Team Consultations

Workshops, teaching sessions and expert consultations that lift your team's skills in AI, software engineering, cloud, data and security, tailored to your own tools and use cases.

Explore training →

Talk to us about AI Agents & Chatbots

Tell us what you need. We reply within one business day and the initial conversation is free of charge.

Discuss your projectBook a consultation or learning sessionsCall 0402 401 584