ClearSignal AI

Inspiration

When you are in an unfamiliar place or speaking another language, understanding the words is only part of understanding the situation. A “free” tour might suddenly require an immediate payment, or a salesperson might discourage you from checking an offer with someone else.

We created ClearSignal AI to provide safety context in uncertain conversations. We want to help travelers, and anyone facing unfamiliar or high-pressure interactions, understand what was said and decide what to ask next.

What it does

ClearSignal AI lets users record a conversation segment, view its English transcript and Spanish translation, and explore potential communication warning signs.

The app analyzes the conversation for possible contradictions, pressure tactics, secrecy requests, and coercive statements. Findings include supporting quotes, explanations, and a suggested neutral question to clarify the situation.

Earlier segments remain available during the session, allowing the analysis to consider the conversation’s context. The current prototype processes each segment after recording stops.

How we built it

We built the frontend with React Native, Expo, and TypeScript, and the backend with Python and FastAPI.

IBM Watson Speech to Text converts recorded audio into a transcript. IBM Watsonx.ai translates the transcript into Spanish and analyzes it alongside earlier conversation segments.

The backend validates the model’s output and retains findings only when their quoted evidence appears in the conversation. The interface presents the transcript, translation, and analysis in expandable cards.

Challenges we ran into

Connecting recording, audio uploads, transcription, translation, and analysis required coordinating several moving parts. We worked through microphone permissions, browser cross-origin restrictions, audio formats, and dependency compatibility.

Another challenge was presenting AI findings responsibly. We needed to distinguish between wording and assumptions about someone’s intentions. We also replaced temporary frontend classification rules with model-generated categories and backend validation.

Accomplishments that we're proud of

We built a working flow from spoken audio to translation and contextual analysis within the hackathon.

We are proud of preserving conversation history for cross-segment analysis and connecting findings to the words that triggered them. Instead of displaying only a warning, ClearSignal AI gives users an explanation and a practical question they can ask.

What we learned

We learned that an AI feature becomes useful through thoughtful integration: recording state, conversation history, structured responses, and clear presentation all matter.

We also learned the importance of defining what our system can support. Our prototype analyzes transcript content; it does not establish a speaker’s honesty or intentions, and it does not reliably interpret vocal emotion or sarcasm.

What's next for ClearSignal AI

We plan to add streaming transcription, support more languages, refine the mobile experience, and evaluate the system across more varied conversations.

We also want to improve how the app communicates uncertainty and explore optional local storage with clear controls. Currently, conversation history stays in memory during the session, while audio and text are sent to IBM services for processing.

Our goal is to give people useful context and better questions when a conversation feels uncertain.

Built With

• expo.io
• fastapi
• ibm
• ibm-watson
• native
• python
• react
• speech
• text
• to
• typescript
• watsonx.ai