A selection of backend, AI, and IoT projects — from hackathon experiments to production systems handling real workloads.
AI-powered conversational agent built during a hackathon. Runs entirely on local LLMs via Ollama with fine-tuned models (UnSloth), uses Qdrant for vector storage, and Pydantic-AI for structured agent orchestration. Awarded 2nd runner-up at the AI Hackathon.
Multi-tenant retrieval-augmented generation platform for a Tokyo-based AI startup. Role: Backend Lead. Architected document ingestion pipelines, hybrid semantic search over ElasticSearch + Qdrant, and LLM-orchestrated Q&A with support for both OpenAI and Anthropic models. Deployed on AWS ECS.
End-to-end observability stack for industrial systems. Ingests real-time telemetry from field devices, stores time-series data in PostgreSQL, and surfaces metrics, logs, and traces through a unified Prometheus + Grafana + Loki + Jaeger dashboard. CI/CD via Jenkins.