AIyuda

What it does

AIyuda guides clients through legal intake in a bilingual conversation, by voice or by text:

Structured start: A short decision tree captures the essentials, such as who the client is, what kind of help they need, and key dates and status.
Adaptive follow-up: AI asks tailored follow-up questions based on the client's answers, so each person's situation is captured without a one-size-fits-all form.
Voice translation: Clients can speak in their own language. AIyuda translates in real time and answers back in natural-sounding speech, so literacy and typing are not barriers.
Plain-language explanations: Legal terms are explained clearly in the client's language as they come up.
Attorney-ready output: The firm receives a standardized summary in English, in a consistent format, ready to review before the first consultation.

Voice streaming: Twilio streams live audio between the client and our backend.
Audio pipeline: Audio moves between services as base64-encoded chunks, which we decode, process, and re-encode for playback to the client.
Voice translation: ElevenLabs powers voice translation, turning the AI's replies into natural speech in the client's language.
Conversation flow: A deterministic decision tree for the opening questions, followed by an LLM-driven follow-up stage [model/provider].
Summary generation: Structured prompting maps the full conversation into a fixed attorney-facing template.

Real-time audio: Keeping the conversation natural meant minimizing delay across the full loop: receiving the client's audio, decoding base64 chunks, translating, generating a reply, and streaming speech back without awkward pauses.
Audio format handling: Twilio and ElevenLabs expect different audio formats, so we had to convert reliably between them mid-conversation.
Balancing structure and flexibility: The decision tree plus AI follow-up gave us both the reliability of a form and the adaptability of a conversation.
Explaining without advising: Plain-language explanations had to stay informational and never become legal advice.
Sensitivity and trust: Clients may share fearful or traumatic experiences, so tone, privacy, and data handling mattered from the start.

Lower latency in the voice loop for more natural conversations.
Attorney validation through pilots with legal aid organizations.
Case-management integration to send summaries directly into firms' existing systems.
Privacy and security hardening for audio and transcripts: encryption, retention controls, and compliance review.