Building ML systems that run where the data is: on the edge, in public services, under real constraints.
- 🔬 I work on LLMs, retrieval-augmented generation and Edge AI, with a focus on ethical deployment in public services.
- ⚡ Recent themes: agent security, on-device inference, and battery / EV health prediction with time-series ML.
- 🧪 I like problems where the model has to survive contact with reality: no GPU, no network, no second chances.
- 📫 serhategeinanc@gmail.com
🏛️ PalisadeRuntime prompt-injection detection and behavioral sandboxing for AI agents. Sits between any agent and its LLM provider. Heuristic filtering plus ML semantic analysis, local-first, no GPU.
|
🎼 fugueMulti-voice agent orchestration for Go. Code-first composition: no YAML, no Markdown, just Go.
|
⭐ RagStarTiny, dependency-free RAG library for Node, Bun, Deno, browsers and edge runtimes. Bring your own embeddings, LLM and vector store. They're just functions.
|
🩺 MedRAGRAG vs. fine-tuned LLMs: a comparative study on the MedQA dataset.
|
🌱 SproutPoint the camera at a plant, get the species. 2,102 classes, fully on-device: no account, no network call, no API key.
|
Battery State-of-Health prediction comparing LSTM, Random Forest and XGBoost on charge/discharge cycle data.
|