Problem

• Product knowledge was scattered across research docs, specs, notes and Jira
• Writing a PRD meant gathering all that context by hand first
• PRDs, briefs and user stories took hours to write from scratch
• Generic AI tools didn't know the team's product or history
• Plans had to be copied into Jira manually
• Product managers were spending more time writing documents than planning products.

Solution Exploration

• Using templates and writing every document by hand
• Using a general-purpose AI chatbot that knew nothing about the team
• AI assistants that work from the team's own documents and connect straight to Jira

How Dezoko Helped

• One home for product knowledge: Teams upload their documents once and can search them by meaning, not just exact keywords.
• AI-drafted planning documents: The AI uses the team's own documents to draft editable PRDs, briefs and user stories.
• Four specialist assistants:
  • One condenses long documents into short summaries.
  • One finds the most relevant information for each task.
  • One drafts PRDs, briefs and user stories.
  • One connects finished plans to the team's Jira backlog.
• No waiting on uploads: Documents are prepared in the background, so uploads finish instantly.
• Secure Jira connection: Jira is connected once for the whole organisation, with access details kept safely locked away.

Governance and Control

• Every change to the AI's instructions is versioned and tracked
• Jira access details are kept in a secure, locked-down store
• Jira access is managed at the organisation level
• The AI runs separately from the main app, so it can be updated on its own

Key Outcomes

• Teams draft planning documents from their own knowledge, in one place
• Uploads finish instantly while documents are prepared in the background
• The AI can be improved without disrupting the main app
• Plans reach Jira without copy-pasting

How We Built It

We built the AI behind maipl around three simple ideas:

Your Team's Knowledge, Ready to Use

• Teams upload their files once and they are saved in the cloud.
• Each document is prepared in the background, so nobody waits for an upload to finish.
• Documents become searchable by meaning, so the AI can find the right context even when the wording differs.

A Team of Specialist Assistants

• Four assistants each handle one job: summarising, finding information, drafting documents and working with Jira.
• Every draft is based on the team's own documents, not generic guesswork.
• Jira is connected once for the whole organisation, so plans flow straight into the backlog.

Built to Improve Safely

• The AI runs separately from the main app, so improvements can ship without disrupting the product people use every day.
• Every change to the AI's instructions is tracked, so it is always clear what changed and when.
• Jira access details are kept in a secure, locked-down store.

Built With

• AI: Google Vertex AI
• Hosting: Google Cloud
• Database: PostgreSQL
• Web App: Next.js and FastAPI
• Integrations: Jira

Business Impact

• AI improvements reach users faster and with less risk
• Uploads are instant, with no waiting around
• Drafts are grounded in the right context from the team's own documents
• Every change to how the AI behaves can be reviewed