3D Conversational Avatar
Autonomous and intelligent 3D conversational virtual assistant. Developed for LotoStudios to innovate customer support.
Company context: Development of a scalable 3D conversational avatar platform for multi-client adoption.
Project Sheet
Orchestration Pipeline
From the user's voice to the avatar's spoken reply, in a single low-latency real-time flow.
User Speech
User speaks to the avatar in the browser
Speech-to-Text
Real-time transcription of audio to text
RAG Ingestion
Scraping, vector embeddings and retrieval from a continuously updated knowledge base
LLM Reasoning
OpenAI orchestrates contextual dialogue and dynamic actions
API & Backend
Cloud-native Node.js: high concurrency and async processing
Polly TTS
AWS Polly generates the avatar's natural voice
3D Avatar
Modular web component: the 3D avatar speaks in the client site
Sample Conversation
Infrastructure & Integrations
As the Backend AI Engineer for LotoStudios, I developed the entire intelligence layer and backend infrastructure from scratch. The main challenge was engineering a low-latency real-time experience, pairing LLM reasoning with user tracking, and orchestrating dynamically generated audiovisual streaming.
Tech & Infrastructure
Key Contributions
Lead backend, integrations, AI services, and DevOps with end-to-end ownership of architecture, implementation, and deployment.
Design and maintain a modular Node.js integration layer deployable across websites with different tech stacks.
Built backend workflows for automated scraping, text chunking, embedding generation, retrieval logic, and dynamic speech synthesis.
Implemented voice generation workflows with Amazon Polly and managed multi-tenant asset delivery on AWS S3.
Operate CI/CD, environment configuration, and infrastructure optimization to support scalability and maintainability.