SpeakUpAndOUT

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for SpeakUpAndOUT

Inspiration

When something important happens, people often have memories, messages, dates, and uncertainty, but no calm way to turn them into a usable record. SpeakUpAndOUT helps a person organize what they already know without surrendering authorship or exposing private material.

What it does

A user can write, paste, or record a short account. SpeakUpAndOUT creates an editable record map containing:

• a voice-preserving summary
• a sourced timeline and reported facts
• uncertainties and contradictions
• evidence to preserve
• steady next steps
• a redacted share draft
• clear "do not touch" warnings

Every factual item points back to numbered source lines. Selecting a reference highlights the supporting words. Suggestions without a direct source remain visibly separate. Nothing is automatically posted or published.

How we built it

The public application is a TypeScript and React vinext project deployed through OpenAI Sites. Its server route calls the OpenAI Responses API with GPT-5.6 Sol and a strict JSON Schema. Returned record maps are validated before reaching the interface. Browser audio is transcribed ephemerally with OpenAI speech transcription.

The app has no account system or workspace database. It does not intentionally retain the account, recording, transcript, or generated map. A clearly labeled deterministic organizer keeps the workflow usable during an API interruption without pretending to be GPT.

How Codex and GPT-5.6 were used

Codex helped build the isolated public architecture, responsive interface, structured-output contract, provenance linking, privacy controls, tests, deployment, original visual assets, and narrated demonstration. Faith Cheltenham retained product direction, voice-preserving behavior, the public/private boundary, and the review-before-sharing rule.

GPT-5.6 Sol performs the record-mapping step. It separates reported material from uncertainty, preserves source references, produces bounded next steps, and drafts redacted public-safe language without making legal conclusions.

Challenges

The hardest design problem was making AI useful without allowing it to overwrite the speaker's account. We solved this with source IDs, strict validation, editable output, explicit route receipts, and a visible distinction between evidence-backed statements and general suggestions.

Accomplishments

SpeakUpAndOUT is public, responsive, testable without an account, and now exercises GPT-5.6 live. It gives users a coherent product experience instead of a black-box summary.

What's next

Future work can add user-controlled encrypted workspaces, richer transcript review, and carefully approved bridges into private evidence tools while preserving the current review and redaction gates.

Built With

• zodhttps://youtu.be/hxtv0rdnifs