Paul Maximilian Magos

Own full-stack development and UI/UX of Hellosmile, an AI voice-agent platform that answers missed calls for European healthcare practices, reducing patient loss during off-hours. Oversee the entire product stack: React/Vercel frontend, FastAPI/ECS backend, LiveKit agents, ElevenLabs TTS, Temporal workflows, Twilio telephony, PostgreSQL/S3 storage, and Grafana/OpenTelemetry observability.

Own full-stack development and UI/UX of Hellosmile, an AI voice-agent platform that answers missed calls for European healthcare practices, reducing patient loss during off-hours. Oversee the entire product stack: React/Vercel frontend, FastAPI/ECS backend, LiveKit agents, ElevenLabs TTS, Temporal workflows, Twilio telephony, PostgreSQL/S3 storage, and Grafana/OpenTelemetry observability.

Built Python microservices for experiment orchestration and data processing, replacing fragile one-off scripts. Introduced MLflow as a unified tracking layer (Docker/Kubernetes), standardising experiment logs/metrics across 4 research entities and reducing debugging time. Automated job scheduling, monitoring, and alerting on a simulation cluster using Proxmox, Docker, and Slurm, which cut manual intervention and improved resource utilization.

Owned a Kafka-based messaging service in a multi-tenant trading platform (Java, Kafka, Kubernetes), improving code by refactoring it with 60% reduction, introducing also structured logging. Prototyped an LLM-based document extraction pipeline (Python, Apache Spark) for financial reports, reaching 75% extraction accuracy while completing the prototype in 20% of the working time. Contributed to CI/CD automation and containerized deployments across environments as well as to internal libraries.

Built end-to-end NLP/multimodal pipelines for tweet retrieval (Twitter API), sentiment analysis (Transformers), image tagging and video transcription, processing 5000 posts/day. Designed and shipped a Python/Vue.js interface that exposed ML results to non-technical internal users.

Stopped early due to career pivot into industry.

Thesis on multivariate time series generation with autoregressive GNNs (PyTorch Geometric). Coursework in ML/DL, parallel & distributed systems, and agile project delivery.

Thesis on analysing skilled migration using Crunchbase datasets. Strong grounding in algorithms, complexity, and software engineering.