Generate Your First POC#

Use the deployed web application to generate and download a small document Q&A POC. Allow roughly 30 minutes for a first run; inference availability, model choice, and validation policy can change the time.

If the service is not deployed yet, start with Installation. This walkthrough confirms the browser workflow; it does not require a source checkout.

Prerequisites#

Check these inputs before opening the creation flow.

  • A deployed POC Factory application and its browser URL.

  • A working administrator or user-saved inference profile, including a compatible embedding model.

  • A validated Cursor key for the default auto generation path, or an intentional llm deployment policy.

  • A user account when authentication is enabled.

Quickstart Steps#

Choose the input path for this first run, then complete the packaging and verification checks.

Follow the browser workflow once to confirm the release can package a POC.

  1. Open the POC Factory URL and sign in if prompted. Open Settings and confirm the effective inference profile is valid. For auto generation, use Validate & Save Cursor Key when the deployment permits user credentials.

  2. Select Create, choose AI Agent, and then choose Proposal to POC. The Generate page opens.

  3. Enter a specific request in Requirements. For example:

    Build a document Q&A assistant for field engineers. It must ingest PDF
and Markdown runbooks, create embeddings, answer questions with citations,
expose a FastAPI API and React UI, include health checks and retrieval
tests, and package the application with Docker Compose. Keep uploaded
documents local.

  4. Leave Domain on auto-detect and leave Enable Human-in-the-Loop Mode off for this first run. The Requirements field needs at least 20 characters; the example provides enough scope for a useful test.

  5. Select Generate POC. Follow the Progress page and its durable Dashboard job entry. The workflow moves through requirements, PRD, optional blueprint selection, architecture, code, tests, validation, documentation, and packaging.

  6. When packaging completes, open the POC from Dashboard. Review validation results, including any skipped gates, and download the archive.

Use the same Create → AI Agent → Proposal to POC flow. Extract the proposal’s implementation requirements into the Requirements field. If you created the proposal through the RFP workflow, use its proposal-to-POC handoff and review the prefilled scope. Review the resulting scope and keep the first job small enough to inspect. Continue through the generation, validation, and download checks described in the first tab.

A successful first run leaves a POC record on Dashboard and provides a downloadable package. If the job pauses or fails, capture its job ID and use Troubleshooting.

Verify the Result#

Confirm that the POC remains on Dashboard, its archive downloads, and the package contains readable instructions, generated source, tests, and deployment files. Inspect contract and validation evidence where supplied. Run the package in the intended test environment rather than assuming every candidate gate ran.

Inference-provider availability and your validation policy affect each run. Review the recorded contract and repair evidence, and check Release notes for changes to generation behavior.

Next Steps#

Continue with review or lifecycle actions after the package is available.

  • Turn on Human-in-the-Loop for a later run when the PRD, blueprint, or architecture needs review. Resume a paused job from Dashboard.

  • Open Accelerate for a completed POC to deploy, evaluate, profile, optimize, monitor, or size it.

  • Read Use POC Factory for application navigation and the generation stages.

  • Read Configure and validate to understand which candidate checks ran in your deployment.

Understand validation

Check what the deployment and generated candidate evidence prove.

configure-and-validate.html
Continue in Accelerate

Explore lifecycle actions and human review for the next run.

../workflows/use-poc-factory.html