David

Senior FullStack Engineer / AI Engineer

Architected and shipped a real-time AI voice agent platform using Python, FastAPI, Django, LiveKit, WebRTC, and Twilio, deployed on GCP (Cloud Run, Cloud Functions, Pub/Sub, Cloud Storage) for scalable low-latency inference.. Integrated Gemini 2.5 Pro and Vertex AI Embeddings into multimodal RAG pipelines combining audio, text, and metadata retrieval, improving grounding accuracy and reducing hallucinations.. Built streaming inference pipelines integrating Whisper/Deepgram STT, ElevenLabs TTS, and multimodal embeddings, delivering sub-second transcription and response generation for interactive applications.. Designed interrupt-driven agent logic with turn-taking and barge-in detection using an event-driven state machine in FastAPI, implementing context-aware state management for natural conversational flow across LLM-powered agents.. Developed end-to-end audio ingestion and preprocessing pipelines using FFmpeg, segmentation, metadata extraction, and validation layers, ensuring reliable downstream inference for both real-time and batch workloads.. Implemented multimodal AI pipelines combining audio, text, and metadata retrieval using RAG, vector search (FAISS/pgvector), and LLM tool-calling, improving grounding accuracy and reducing hallucinations.. Built real-time WebSocket APIs for bidirectional audio streaming, model inference, and agent orchestration, supporting concurrent sessions under strict latency budgets.. Implemented multi-agent orchestration using LangGraph and LlamaIndex, enabling autonomous workflows, multi-step reasoning, and tool-calling pipelines for complex conversational and retrieval tasks.. Built batch inference pipelines for large-scale evaluation, dataset generation, and offline scoring, enabling continuous improvement of speech and retrieval models.. Delivered reproducible deployments using Docker, GitHub Actions, AWS S3, and internal CI/CD pipelines, reducing deployment time and ensuring consistent releases.. Collaborated with speech scientists, clinicians, and product teams to validate outputs, refine system behavior, and improve prediction accuracy for clinician-facing UIs through structured evaluation and user testing.. Leveraged AI coding tools (Claude Code, GitHub Copilot) to accelerate development, improve code quality, and streamline debugging across AI and full-stack systems.. Mentored junior engineers through architecture reviews, pair programming, and AI workflow training, improving team velocity and raising engineering standards.. Led cross-functional initiatives coordinating engineering, product, and research teams to deliver new AI features such as multimodal embedding services within tight deadlines.