Inspiration
The idea for Lwazi was born at the intersection of law and technology, sparked during the Harvard Law School Executive Education programme in September 2024. Conversations with thought leaders such as Professor David Wilkins and Robert Mahari made it clear: law firms that fail to innovate will be left behind. At Hadar Incorporated, we saw an opportunity to lead the charge in South Africa by reimagining client service through conversational AI.
Lwazi means “knowledge” in Zulu. She embodies our vision for Law Firm 2.0—a firm that is client-centric, efficient, and unafraid of tech disruption. We wanted to build more than a chatbot; we wanted to create a voice-enabled, real-time assistant that simplifies legal interactions and delivers real value.
What it does
Lwazi is South Africa’s first voice-based AI agent built specifically for Property Law. She speaks, listens, understands, and guides users through common legal processes. She: • Answers client calls directly (+27 060 042 8851) • Collects contact details and matter summaries • Clarifies legal issues through structured conversation • Provides jargon-free explanations of legal steps • Summarises the call and sends the information to the Hadar legal team for follow-up • Lives both on our website (www.hadarinc.co.za) and as a phone-accessible AI
Lwazi is built with Retrieval-Augmented Generation (RAG), meaning her responses are accurate, contextual and based on Hadar Inc.’s own legal database—not hallucinations.
How we built it
We used a layered tech stack integrating: • ElevenLabs for real-time voice synthesis and recognition • OpenAI/GPT-4 and Gemini for LLM-backed natural language understanding • Make.com to orchestrate workflows • Twilio for call handling and SMS • Google Sheets for data storage
Custom lexicons were uploaded to ElevenLabs to ensure Lwazi speaks in a South African tone, pronouncing names and terms in a locally accurate way. The backend is structured to summarise conversations, extract name/phone/email/issue data, and pass that directly to our legal team.
Challenges we ran into • Pronunciation accuracy: ElevenLabs’ default pronunciation often sounded American or robotic. We overcame this by building a custom alias dictionary. • Data capture reliability: Ensuring Lwazi correctly parses phone numbers and emails required regex validation and fallback error handling. • Voice latency: Balancing real-time responses with clarity meant tuning latency, speed, and stability in ElevenLabs. • Legal risk boundaries: We had to ensure that Lwazi never provides formal legal advice, only helpful guidance.
Accomplishments that we’re proud of • Launching South Africa’s first voice-enabled legal AI agent • Integrating real-time phone call handling into a seamless automation pipeline • Preserving a distinctly South African tone in both accent and cultural context • Creating a scalable, modular system that can be updated and trained with new legal data • Receiving strong early feedback from users who find the experience friendly and informative
What we learned • The user experience must come first. No one wants a robot that sounds like a robot. • Voice is a powerful medium for the law. Clients respond better when they can talk rather than type. • South African users want simplicity, clarity, and respect in every interaction. • AI works best as a collaborator, not a replacement, for human professionals.
What’s next for Lwazi Prop Law Specialist • Client Matter Updates: In Phase 2, Lwazi will give existing clients updates on their matters (e.g. court status, document readiness). • Multilingual Support: Expanding into isiZulu, isiXhosa and Afrikaans for more inclusive access. • SMS Summaries: After each call, the user will receive a text summary with a follow-up link. • API Integrations: Connecting Lwazi to CRMs and court data for better automation. • Visual UI: Launching a WhatsApp and Web-based Lwazi with voice-to-text backup.
We believe this is just the beginning. Lwazi is not just a tool, she is a promise—a promise that the law can be more accessible, more human, and more helpful for all South Africans.
Built With
- bolt
- elevenlabs
- googlesheets
- gpt-4.1
- make.com
- twilio

Log in or sign up for Devpost to join the conversation.