Conversational AI developer

What does a Conversational AI developer do?

A conversational AI developer builds chatbots and voice or text assistants that talk with customers, staff or citizens.

The developer connects large language models (LLMs) to company data, writes prompts, and tests that answers are correct and safe.

In Singapore, the role sits in a technology, data or digital team. You find it in banks, insurers, telcos, government agencies, e-commerce firms, and agencies that build bots for clients.

You work with product managers, customer service teams, UX writers, data engineers and security staff. Compliance and legal teams also review your work when the bot handles personal data or regulated advice.

What the work involves

• Write and refine prompts and system instructions so the assistant answers in the right tone and scope.
• Build retrieval pipelines that let the assistant find answers in policy documents, help articles and product data.
• Connect the assistant to company systems through APIs, such as order lookup or appointment booking.
• Read chat logs, find where the bot failed, and fix the cause in prompts, data or code.
• Run test sets before each release and report accuracy, escalation rates and risks to the team.

Pay in Singapore

Back to top

Skills a Conversational AI developer needs

Back to top

Conversational AI developer job description

We are hiring a conversational AI developer to design, build and improve AI assistants for our customers and staff in Singapore.

You will work in Python with LLM APIs, retrieval and agent tools. You will join a small team, work with product and customer service, and take assistants from prototype to live use.

Responsibilities

• Design conversation flows and decide when the assistant answers, asks a question or hands over to a person.
• Write, test and version prompts and system instructions.
• Build retrieval-augmented generation (RAG) pipelines over internal documents and knowledge bases.
• Integrate LLM APIs and company systems through secure APIs.
• Build agent tools that let the assistant take actions, with checks on inputs and outputs.
• Create test sets and measure answer quality, safety and handover rates before and after release.
• Monitor live conversations, review failures and ship fixes.
• Apply personal data, security and content rules set by the organisation.
• Write clear documentation and explain limits to business teams.
• Track cost and response time of model calls and tune them.

Requirements

• Degree or diploma in computer science, data science, engineering or a related field, or equivalent hands-on project experience.
• Typically 1 to 3 years building software, with at least one chatbot or LLM project you can show.
• Solid Python and experience with REST APIs and JSON.
• Hands-on use of LLM APIs, prompt design and RAG, including vector search and embeddings.
• Experience with testing, logging and cloud deployment, plus clear written English for working with non-technical teams.

Back to top

Conversational AI developer career path

Most conversational AI developers start as junior software or data staff and move towards senior engineering, technical leadership or product roles.

Titles vary by employer.

1. Junior conversational AI developer

   Entry

   Builds and tests bot features under guidance, writes prompts, fixes bugs and reviews chat logs.

2. Conversational AI developer

   Mid

   Owns a bot or a major feature, builds RAG pipelines and integrations, and sets up testing.

3. Senior conversational AI engineer

   Senior

   Designs the architecture for several assistants, sets quality and safety standards, and mentors others.

4. Lead conversational AI engineer

   Lead

   Leads a team, plans delivery with product and business owners, and manages vendor and model choices.

5. Head of conversational AI

   Head

   Sets the direction for AI assistants across the organisation, including budget, risk and hiring.

The titles are ones commonly used in Singapore; employers may name the levels differently.

Related jobs and their pay

Back to top

Conversational AI developer interview questions

We wrote these questions from the skills this job needs. They are not an official question bank.

1. Design a RAG pipeline for a bank's customer FAQ assistant. What are the main steps?

   What a good answer covers

   • Cleans and chunks documents with sensible chunk size and metadata
   • Chooses an embedding model and a vector store, and explains why
   • Retrieves, then reranks or filters results before generation
   • Makes the answer cite its source passages
   • Plans a way to refresh content when policies change

2. How do you reduce hallucinations in an assistant that answers from company documents?

   What a good answer covers

   • Grounds answers in retrieved text and tells the model to say when it does not know
   • Improves retrieval quality before changing the prompt
   • Adds citations and answer checks against the source
   • Tests with questions that have no answer in the documents
   • Sends low-confidence cases to a person

3. Walk us through how you structure a system prompt for a support assistant, and how you keep changes under control.

   What a good answer covers

   • Separates role, scope, tone, rules and output format
   • Gives short examples of good and bad replies
   • Stores prompts in version control with a change log
   • Runs a fixed test set after each change
   • Keeps out rules that the code should enforce

4. An LLM API call times out or returns a rate-limit error during a live chat. What does your code do?

   What a good answer covers

   • Retries with backoff and a sensible limit
   • Streams output so the user sees progress
   • Falls back to a smaller model or a polite message
   • Logs the error with a request ID for debugging
   • Sets timeouts and monitors cost and latency

5. How do you manage conversation memory when a chat grows beyond the model's context window?

   What a good answer covers

   • Keeps recent turns and summarises older ones
   • Stores key facts, such as the customer's request, in structured form
   • Drops sensitive details that are not needed
   • Tests that the bot still follows earlier instructions
   • Explains the cost and accuracy trade-offs

6. Describe how you would let an assistant check an order status through tool calling.

   What a good answer covers

   • Defines a clear tool schema with typed inputs
   • Validates arguments before calling the order API
   • Checks the user is allowed to see that order
   • Handles errors and empty results with a clear message
   • Logs each tool call for audit

7. How would you use Python to handle many chat sessions at once without slow replies?

   What a good answer covers

   • Uses async code for network-bound model and API calls
   • Keeps session state in a store such as Redis instead of in memory
   • Limits concurrent model calls and queues the rest
   • Measures latency under load before release
   • Keeps secrets out of code

8. How do you measure whether a customer support assistant is ready to launch?

   What a good answer covers

   • Builds a test set from real past questions
   • Scores accuracy, safety and tone, with human review of samples
   • Tracks handover rate and resolved-without-agent rate
   • Includes tricky and hostile inputs
   • Sets clear pass marks agreed with the business

1. Two days after launch, the bot quotes a refund rule that no longer exists. What do you do?

   What a good answer covers

   • Contains the issue first, for example by blocking that topic or adding a warning
   • Finds the cause, such as old documents in the index
   • Fixes the data and adds a test for that question
   • Checks logs for other customers who got the wrong answer
   • Tells the business owner what happened and what changed

2. A manager wants to launch in two weeks and skip testing. How do you respond?

   What a good answer covers

   • Explains the specific risks in business terms
   • Offers a smaller launch scope or a limited pilot
   • Proposes a quick test set that fits the timeline
   • Agrees on what the bot will leave unanswered at first
   • Documents the decision

3. Customers say the bot loops and will not let them reach a person. How do you redesign the handover?

   What a good answer covers

   • Defines triggers such as repeated failure, anger or sensitive topics
   • Passes the chat summary to the agent so customers do not repeat themselves
   • Shows the handover option clearly in the chat
   • Handles out-of-hours cases
   • Tracks handover outcomes to improve the bot

1. Tell us about a time you explained the limits of an AI system to a non-technical colleague.

   What a good answer covers

   • Names the situation and the audience
   • Uses everyday examples and avoids jargon
   • Shows the colleague's concern was taken seriously
   • Describes what was agreed afterwards

2. Describe a project where user testing showed your first design was wrong. What did you change?

   What a good answer covers

   • Shows how the problem was found, using logs or user feedback
   • Admits the mistake without blame
   • Explains the change and how it was checked
   • Shares what was learned for later work

1. Your bot stores chat transcripts that include NRIC numbers and phone numbers. What do you raise before launch?

   What a good answer covers

   • Collects only the personal data the bot needs
   • Tells users how their data is used and gets consent where required
   • Masks or removes NRIC numbers and other identifiers in logs
   • Sets retention limits and access controls
   • Checks where the model provider processes and stores data, and involves the data protection officer

2. Customers write in English mixed with Singlish, Mandarin and Malay. How do you handle this in your assistant?

   What a good answer covers

   • Collects real examples of local phrasing for testing
   • Tests the model on code-switched messages
   • Adds clarifying questions when intent is unclear
   • Decides which languages to support and states this to users
   • Reviews failures regularly with native speakers

Common questions about a Conversational AI developer job

Sources