What’s Next

Forms Lab demonstrates the foundation. Here’s where it goes from here:

Planned Capabilities

• Conversational form filling — Carlos completes complex sections through dialogue with an LLM agent (Story #9)
• LLM-assisted refinement — Maya uses conversation to iteratively improve the data model (Story #8)
• Advanced evaluation — LLM-as-Judge scoring for richer, semantic evaluation of extraction quality

Open Questions

• How well does extraction generalize across form types (tax forms, benefit applications, regulatory filings)?
• What’s the right balance between automatic extraction and human refinement?
• Can the evaluation framework itself be used to improve extraction prompts automatically?

The Bigger Picture

Government forms represent one of the highest-volume, highest-impact touchpoints between agencies and the public. Making these forms easier to create, more accessible to fill out, and more reliable to process is a meaningful application of LLM technology.

See: All Stories | Architecture