Parley

Inspiration

Learning through talking and interaction

What it does

Our application functions as an aid in speech and learning, especially focused at those who have a verbal learning style. The chat feature, either through text or spoken input, allows the user to interact with their content at the level and pace that is suited to them. The AI interface offers different styles of responding and variable complexity, based on the conversation history and the needs of the user.

How we built it

This project blends a web-based conversational UI with a realtime server that orchestrates speech input, speech output, and large-language-model reasoning: the browser shell uses modern component styling plus built-in speech APIs to capture utterances, stream interim transcriptions, maintain chat history, and relay each turn over a persistent socket; the server pairs that connection with a session manager that can ingest study materials, maintain dialogue context, route user turns through a configurable language-model interface, and—when enabled—pipes results through cloud speech services so responses are spoken aloud.

Challenges we ran into

From the start of the day on Friday, we ran into multiple challenges, most notably dependency issues.

We started off very quickly on the implementation of a conversational agent.. Building a frontend UI became increasingly more difficult as we kept incorporating more features. The primary one we wanted to include was a seamless vocal component which senses when the speaker finishes talking to call the inference component.

We held ambitious feature goals for our prototype. We had initial difficulty finding a frontend-backend workflow which allowed us to work together in sync.

Accomplishments that we're proud of

Conversation Agent implementation through LangChain. Seamless socket chat implementation with Qwen3 via Cerebras Inference. A user-friendly chatbot interface. Effective prompting. Real-time text-to-speech using custom vocals and live input

What we learned

Avoid ambitious feature engineering. Start with a simple product and be able to build from there. Some of the development into designing an LLM and both frontend and backend Working concurrently and coherently as a team and ensuring updates are synced with the team.

What's next for Parley

Evaluation Benchmarks

• Test how well people retain information when using Parley versus studying traditionally or using general commercial LLMs like ChatGPT. Shaping the User Experience
• Find use cases our users enjoy
  • Studying for exams
  • Interview Preparation
  • Self-study and Entertainment
• Refining LLM prompting to support use cases or needs.
• Collect and Analyze User Feedback Implement new features
• PDF upload, be able to interact with Parley on your uploads

Built With

• python
• vercel