Skip to main content

How Chat Aid Generates Answers

Learn how Chat Aid processes your questions and generates accurate, contextual answers using your company's data.

Overview​

Chat Aid uses advanced AI technology called Retrieval-Augmented Generation (RAG) to answer questions using your company's trained data. This approach combines powerful search capabilities with AI language models to provide accurate, source-backed answers.

The Answer Generation Process​

1. Question Understanding​

When you ask a question, Chat Aid:

  • Detects the language of your question
  • Extracts the core question from conversational context
  • Understands the intent behind your query
  • Identifies any special search filters (like specific data sources or teams)

2. Smart Search Strategies​

Chat Aid uses multiple intelligent search strategies to find the most relevant information:

Indexed Data Search (Try 1-3):

  • Vector Search: Converts your question into a mathematical representation to find semantically similar content
  • Hybrid Search: Combines semantic search with keyword matching for better precision
  • Conversational Context: Uses conversation history to understand follow-up questions

Live Data Search (Try 4 - Agentic Flow):

  • If indexed search doesn't find satisfactory results, Chat Aid can automatically query live data from your connected applications
  • Currently supported: Jira (querying tickets, status, assignees, and more in real-time)
  • Returns real-time data directly from Jira's API when needed
  • Additional connectors may be enabled for your organization upon request

Multiple Attempts:

  • Chat Aid automatically progresses through search strategies until a good answer is found
  • Ensures the best possible answer by combining indexed and live data sources

3. Intelligent Ranking​

Once relevant documents are found, Chat Aid:

  • Retrieves up to 500 candidate documents from your data
  • Reranks them using advanced AI models to find the most relevant ones
  • Selects the top 10 most relevant chunks to use for generating an answer

This reranking step is crucial for answer quality, as it uses sophisticated language models to understand which documents truly answer your question.

4. Answer Generation​

With the most relevant information identified, Chat Aid:

  • Synthesizes information from multiple sources
  • Generates a coherent answer using the AI model based on your Chat Aid plan
  • Cites the sources used to create the answer
  • Maintains your persona's tone if configured
  • Answers in the same language as your question

Quality Assurance:

  • Answers are based only on your trained data, not external knowledge
  • Sources are evaluated for relevance before inclusion
  • The AI is instructed to say "I could not find an answer" rather than guess

5. Fallback Strategies​

If Chat Aid cannot find a good answer in your company data:

Internet Fallback (Optional):

  • If enabled, Chat Aid will search the internet for general knowledge answers
  • Uses a general AI model for broad public knowledge
  • Clearly indicates when an answer comes from the internet vs. your data

No Answer Response:

  • If no good answer is found and internet fallback is disabled, Chat Aid will respond: "I could not find an answer to your question."
  • This prevents hallucination or guessing

What Chat Aid Does Well​

Chat Aid excels at:

✅ Knowledge Retrieval​

  • Finding specific information in your documents
  • Synthesizing information from multiple sources
  • Answering "what," "how," and "why" questions
  • Providing context from related documents

✅ Live Data Access (Jira)​

  • Querying real-time Jira data when indexed search isn't sufficient
  • Retrieving current ticket status, assignees, and details
  • Looking up specific tickets by ticket ID (e.g., PROJ-123)
  • Automatically falling back to live Jira API when needed

✅ Source-Backed Answers​

  • Citing exact sources for every answer
  • Linking back to original documents for verification
  • Maintaining accuracy by staying grounded in your data
  • Combining indexed and live data sources when relevant

✅ Conversational Memory​

  • Understanding follow-up questions in the same conversation
  • Maintaining context across a thread
  • Clarifying ambiguous questions based on conversation history

Current Limitations​

❌ Aggregate Queries​

Chat Aid cannot answer questions requiring aggregation across all your data, such as:

Examples of Unsupported Queries:

  • "How many total open tickets are assigned to me in Jira?"
  • "What's the sum of all Q4 revenue across all accounts?"
  • "List all employees who joined in the last 30 days"

Why This Limitation Exists:

  • Chat Aid uses the top 10 most relevant documents to generate answers
  • It does not have access to your entire database for computation
  • It's designed for knowledge retrieval, not database querying

Workarounds:

  • Use the native application's reporting features (e.g., Jira reports, Salesforce dashboards)
  • Ask questions about processes or policies instead: "How do I find my open Jira tickets?"
  • Consider using Actions to query live data from connected apps

Query Filters and Search Control​

Chat Aid supports special query filters to control how your questions are answered:

Available Filters​

[wiki] - Company Data Only:

[wiki] What is our vacation policy?

Searches only your indexed company data sources, skipping live queries and internet search.

[agent] - Live Data Search:

[agent] Show me my open Jira tickets

Forces Chat Aid to query live data from your connected applications instead of (or in addition to) indexed data.

[ai] - Internet Search:

[ai] What is the current weather in New York?

Uses internet search for general knowledge questions (if internet fallback is enabled for your organization).

[action] - Execute Tasks:

[action] Create a Jira ticket for this bug

Routes your request to the Actions feature to execute tasks rather than just answer questions.

When to Use Query Filters
  • Use [wiki] when you want to ensure the answer comes only from your company data
  • Use [agent] when asking about current status, tickets, or CRM data that changes frequently
  • Use [ai] for general knowledge questions unrelated to your company data
  • Filters work on all platforms: Slack, Microsoft Teams, and the web dashboard

How Live Data Search Works​

When you use [agent] or when Chat Aid automatically triggers live data search (Try 4) for Jira:

  1. Jira Detection: Chat Aid detects Jira-related queries or ticket references
  2. API Queries: Intelligent agents query your connected Jira instance via the Jira API
  3. Parameter Extraction: Extracts search parameters (ticket ID, status, assignee, project, priority, etc.)
  4. Result Synthesis: Results from live Jira API are combined with indexed data when relevant
  5. Source Citations: Live Jira data is cited alongside indexed documents

Example Flow:

You: [agent] What's the status of PROJ-456?

Chat Aid:
1. Detects Jira ticket reference (PROJ-456)
2. Queries Jira API for ticket PROJ-456
3. Retrieves real-time status, assignee, comments, priority
4. Generates answer with live data and citations

Supported Jira Query Parameters:

  • Ticket ID (e.g., PROJ-123)
  • Assignee names
  • Ticket status (In Progress, Done, etc.)
  • Priority (High, Medium, Low)
  • Project name
  • Ticket types (Bug, Feature, Task, etc.)
  • Date ranges (created, updated, completed)
  • Tags and labels

Advanced Customization​

Personas​

Configure personas to customize how Chat Aid responds:

  • Adjust tone and style
  • Add domain-specific instructions
  • Define response formats

Response Settings​

Control where and how Chat Aid responds:

  • Choose which data sources to search
  • Configure AI fallback behavior
  • Set channel-specific overrides

Teams​

Create department-specific knowledge bases:

  • Isolate sensitive data by department
  • Provide role-specific answers
  • Maintain data access control

Have questions about how Chat Aid works?

Need Help?

Our support team is here to help you get the most out of Chat Aid.