AI Customer Service Knowledge Base & Reply Assistant Developer

We are looking for an experienced AI developer to build a reliable internal customer-service knowledge and training system for a growing fitness equipment company.

The owner personally handled nearly all customer service for many years. As we add customer-service representatives, we need to transfer that knowledge into a system that helps employees find accurate answers, learn our products, and respond consistently.

We currently use Zoho Desk and have already begun creating knowledge-base articles from historical customer emails and support tickets. However, we need a much more complete and maintainable system.

The finished system should help customer-service representatives answer questions accurately without depending on the owner for every situation.

What We Want to Build

1. Internal Customer-Service Reply Assistant

We want a simple place where a customer-service representative can enter or paste a customer’s question and receive:

• A suggested response written in our normal customer-service tone.
• The relevant product information or company policy used to create the response.
• Links to the supporting knowledge-base articles, setup pages, manuals, or approved source material.
• A confidence indicator or warning when the information may be incomplete.
• Clear instructions to escalate the question when the system does not have enough verified information.
• A response that the representative can review, edit, and copy into the customer email.

The initial system should assist representatives rather than automatically sending replies to customers. Human review and approval are required.

2. Continuously Updated Knowledge Base

We want a system that regularly reviews new customer-service emails or Zoho Desk ticket threads and identifies useful information that is not already covered in the knowledge base.

The system should:

• Review new ticket conversations on an ongoing basis.
• Exclude routine order-status and shipping-update conversations when they do not provide reusable product knowledge.
• Identify repeated and unique product questions.
• Compare new questions against existing knowledge-base content.
• Avoid creating duplicate or nearly identical articles.
• Combine closely related questions when that produces a more complete article.
• Give more weight to recent approved information when older answers conflict.
• Draft proposed questions and answers for human review.
• Show the source tickets or messages used to create each proposed article.
• Allow an authorized person to approve, edit, reject, or merge proposed articles.
• Track when an article was created, updated, approved, or replaced.
• Publish or export approved content in a format compatible with Zoho Desk.
• Detect when existing articles may be outdated because newer support responses contain different information.

3. Customer-Service Training System

We also want this system to support employee training by:

• Organizing knowledge by product and topic.
• Separating installation instructions from troubleshooting.
• Creating product-specific learning paths.
• Providing searchable procedures, policies, and escalation guidelines.
• Helping turn undocumented owner knowledge into SOPs and decision trees.
• Identifying knowledge gaps based on questions employees and customers continue to ask.
• Tracking which topics create the most confusion or require owner escalation.
• Potentially generating quizzes or practice customer scenarios from approved information.
• Helping management identify where additional training materials are needed.

4. Reporting and Quality Control

The system should help us understand:

• Which customer questions occur most often.
• Which questions do not have approved answers.
• Which articles representatives use most often.
• Which suggested responses representatives edit significantly.
• Which subjects are regularly escalated.
• Where the AI produces low-confidence answers.
• Whether response consistency and first-response time are improving.
• Which knowledge articles may need review or updating.

Current Systems and Sources

Relevant systems and materials may include:

• Zoho Desk tickets and knowledge-base articles.
• Historical email exports.
• Existing Q&A spreadsheets.
• Shopify product pages.
• Product specifications and compatibility information.
• Setup pages and videos.
• Installation manuals.
• Google Docs and Google Sheets.
• Internal policies and procedures.

The contractor should recommend the best architecture after reviewing our current systems. We are open to APIs, databases, retrieval-based AI systems, automation platforms, and custom interfaces, but the final solution must be practical and maintainable.

Important System Requirements

Accuracy is more important than producing an answer to every question.

The system must:

• Use approved sources as the basis for responses.
• Provide traceability back to its sources.
• Avoid inventing product specifications, compatibility information, policies, or troubleshooting instructions.
• Clearly state when there is not enough information to answer.
• Route uncertain, aggressive, sensitive, or unusual situations to a human.
• Protect customer information and limit access appropriately.
• Avoid exposing information from one customer’s ticket to another customer.
• Maintain logs and version history.
• Allow knowledge to be corrected without rebuilding the entire system.
• Include testing and evaluation before representatives rely on it.

We do not want a basic chatbot that simply connects to a folder of documents. We need a carefully designed knowledge workflow that remains accurate as products and policies change.

Required Experience

• Proven experience building AI-powered customer-support or internal knowledge systems.
• Experience with retrieval-augmented generation, document search, embeddings, or comparable knowledge-retrieval methods.
• Experience connecting help-desk platforms, email systems, databases, and APIs.
• Experience creating human approval workflows.
• Experience with document classification, deduplication, content versioning, and source tracking.
• Knowledge of AI evaluation, hallucination prevention, confidence handling, and escalation rules.
• Strong understanding of customer-data privacy and access controls.
• Ability to create a straightforward interface for nontechnical employees.
• Strong documentation and training skills.

Experience with Zoho Desk, Shopify, ecommerce customer service, product support, or technical product knowledge bases is strongly preferred.

Expected Deliverables

The exact architecture will be finalized during discovery, but expected deliverables may include:

• Current-system and data-source assessment.
• Recommended architecture and implementation plan.
• Working prototype.
• Approved-source knowledge database.
• Customer-service question and response interface.
• Ongoing ticket-review and article-drafting workflow.
• Duplicate detection and article-update workflow.
• Human review and approval system.
• Source citations and confidence handling.
• Zoho Desk integration or Zoho-compatible export process.
• Reporting dashboard.
• Testing and evaluation framework.
• Security and access documentation.
• Administrator instructions.
• Customer-service representative training.
• Ongoing maintenance plan.

How to Apply

Please answer every question below. Generic proposals or descriptions of basic chatbots will not be considered.

1. Provide examples of customer-service assistants, internal knowledge systems, or retrieval-based AI tools you personally built. Explain your exact role and the measurable result. 2. How would you process new Zoho Desk tickets, determine whether they contain new knowledge, and avoid creating duplicate articles? 3. How would you handle a situation where a newer customer-service response conflicts with an older knowledge-base article? 4. How would you ensure every suggested customer reply can be traced to approved source material? 5. What would the system do when it cannot find enough reliable information to answer? 6. How would you prevent customer-identifying or private information from entering knowledge articles or appearing in another customer’s response? 7. How would you test whether the assistant’s answers are accurate before customer-service representatives begin using it? 8. Which parts would you build with existing platforms, and which parts would require custom development? 9. Describe the user interface you would recommend for a customer-service representative. 10. What information and access would you need from us during the initial discovery phase? 11. Provide a preliminary timeline and budget range for discovery, a prototype, and a production-ready version. 12. Are you an independent contractor or part of an agency? Who would perform the actual work? 13. What is your time zone and general availability? 14. Begin your proposal with “CS Knowledge System” so we know you read the complete posting.

Important Communication Requirement

Please communicate with us through Upwork only. Do not call our company, email us directly, submit website contact forms, or contact our customer-service channels. Applicants who contact us outside the job posting will not be considered.