streaming-chat-agent
Token-streaming chat agent with server-sent events, tool calls, and short-term conversation memory. A playground for real-time LLM UX — backpressure, retries, and clean cancellation.
Stack
Un aperçu de récentes missions clients — ce qui a été construit, la stack utilisée et le rôle tenu.
Token-streaming chat agent with server-sent events, tool calls, and short-term conversation memory. A playground for real-time LLM UX — backpressure, retries, and clean cancellation.
Structured field extraction from unstructured text. Prompts an LLM to pull named entities, normalizes them against a schema, and emits clean JSON for downstream pipelines.
Compact retrieval-augmented generation demo: chunking, embeddings, vector search, and grounded LLM responses. Built as a learning surface for retrieval patterns and prompt design.
Research on heterogeneity, attacks, and robustness in federated learning. PyTorch + Flower framework, evaluating privacy attacks and defenses on cross-device FL.
PHP backend for an AI symptom-triage interface. Conversational endpoint, prompt structuring, and a clean REST surface for any frontend to plug into.
Strava integration that turns ride telemetry into structured coaching feedback. OAuth flow, activity ingestion, and rule-based training cues.
Web app teaching Modern Standard Arabic to Korean speakers — alphabet, vocabulary, daily phrases, and beginner grammar, with a quiz layer. PHP-only, no database; lessons authored as plain arrays.
Greenhouse monitoring system in C++ — environmental sensing, control loops, and configuration handling for an automated cultivation setup.
Apprentissage fédéré, attaques sur la vie privée et robustesse — travaux en cours au SKKU InfoLab.