Quickstart
Go from zero to your first AI-generated insight in under 5 minutes.
Overview
This guide walks you through the core Praxiom AI workflow end-to-end: creating a workspace, uploading research, synthesizing insights, generating recommendations, drafting a document, structuring it into blocks, and generating execution tickets. By the end you will have a working product loop running on real data.
Prerequisites
- A Praxiom AI account at praxiomai.xyz
- An active access code (provided during onboarding or by your team admin)
- At least one research file ready to upload (PDF, audio, CSV, or text)
Step-by-step
Create a workspace
After signing in, hit New Workspace. Fill in your product name, a short description, target users, and current stage. This context helps the AI tailor its analysis to your product.
The API equivalent is POST /api/workspaces with fields name, product_name, product_description, target_users, and current_stage.
Upload your research
Navigate to Research and click Upload. Drag in your files — interviews, support tickets, survey exports, or any document that captures user voice. Praxiom accepts PDFs, audio/video, images, CSVs, and plain text up to 100 MB per file.
Each upload hits POST /api/research/upload as multipart form data. You must supply a workspace_id, source_type (one of interview, support_ticket, survey, usage_data, feedback, other), and a title.
Background processing starts immediately. The source moves through pending to completed once text extraction or transcription finishes.
Synthesize insights
Once your sources show Completed, select the ones you want to analyse and click Synthesize. The AI reads every source, cross-references themes, and extracts structured insights with direct quotes, severity ratings, and frequency counts.
Under the hood this calls POST /api/research/synthesize with your workspace_id, an array of source_ids, and a synthesis_type. Synthesis typically takes 15-60 seconds.
Generate recommendations
Open the Insights panel, select the insights you care about, and hit Generate Recommendations. The agent analyses patterns across your selected insights and produces 3-7 buildable feature recommendations, each scored by impact and effort.
This triggers POST /api/recommendations/generate with your workspace_id and insight_ids. You can optionally pass focus_areas to steer the output.
Draft a document
Pick an accepted recommendation and choose a document type — PRD, Technical Spec, Decision Memo, User Stories, Roadmap Update, or Release Notes. Praxiom drafts a complete document grounded in your research, with inline citations and quotes.
The endpoint is POST /api/documents/draft. Supply your workspace_id, document_type (e.g. prd, spec, memo), and optionally a recommendation_id and user_instructions.
Structure into blocks
Open your document in the PM IDE to see it decomposed into typed blocks — problem statements, personas, user stories, metrics, risks, and more. Each block can be individually refined:
- Expand — Flesh out a block with deeper detail using its type-specific contract
- Challenge — Get an adversarial critique with strengths, weaknesses, and improvement suggestions
- Simulate — Explore real-world implications and edge cases
You can also use POST /api/intent/parse to decompose raw input into a structured block skeleton automatically.
Generate execution tickets
Select document blocks and generate scoped engineering tickets with POST /api/execution/generate. Each ticket includes a title, description, acceptance criteria, priority, and effort estimate. Review and edit tickets, then push them to GitHub, Linear, or Jira with POST /api/execution/push.
You can also drive every step from the chat agent. Just open a conversation and ask: "Synthesize my interview data", "Draft a PRD for the onboarding recommendation", or "Generate tickets from my PRD."
Choosing a Model
Praxiom supports three Claude model tiers, each with configurable effort levels:
| Model | Best For | Effort Levels |
|---|---|---|
| Haiku | Quick lookups, simple queries | fast, thorough |
| Sonnet | Balanced analysis, most workflows | fast, thorough, deep |
| Opus | Complex synthesis, deep research | fast, thorough, deep |
Higher effort levels enable extended thinking for more thorough analysis at the cost of additional credits. See the Models & Effort Levels guide for details.
Key Concepts
| Concept | Description |
|---|---|
| Workspace | The container for all your product data — sources, insights, recommendations, documents, and blocks. |
| Source types | interview, support_ticket, survey, usage_data, feedback, other |
| Processing status | Every uploaded source goes through pending → completed (or failed). Only completed sources can be synthesized. |
| Document blocks | Typed content sections within documents: problem, persona, solution, metrics, user_story, edge_case, risk, timeline, free_text |
| Execution tickets | Scoped engineering tickets generated from document blocks, pushable to GitHub, Linear, or Jira |
What's Next
Learn how authentication works and how to call the API directly in the Authentication guide.
Was this helpful?