AI Engineer Intern — RAG Core (UniAI Team)
Universitas Hasanuddin (DSITD)
Technical Role•Feb 2026 – Present
Makassar, South Sulawesi
Building the data and RAG core of UniAI, an institutional academic chatbot for Hasanuddin University, as part of Team 2 (Data & RAG Core). Developed an end-to-end retrieval pipeline using Qwen3-Embedding for dense retrieval, a BGE-Reranker, a Qdrant vector database, and Redis semantic caching to cut latency and inference cost. Designed a 6-layer safety and routing pipeline (keyword filter, Llama Guard, IndoBERT intent classifier, private API handler, RAG pipeline, output filter) and a three-tier OCR ingestion strategy (PyMuPDF, Tesseract, Qwen3-VL). Deployed and served Qwen3-VL-8B-Instruct via vLLM behind FastAPI and Nginx on an NVIDIA L40S GPU, load-testing toward 500 concurrent users.
- RAG
- LLM
- vLLM
- Qdrant
- FastAPI
- Python
- Embeddings
- Redis
- Docker