AI Engineer focused on Prompt Engineering, RAG pipelines, and LLM application development — building evaluation-driven AI systems from design to deployment.
B.Eng. in Automation & Control Engineering Technology (HCMUTE) | GPA: 3.4/4.0
- Prompt Engineering — Design specialized prompts (system/role, structured JSON output, temperature tuning, context grounding) and iterate based on evaluation metrics
- RAG Pipeline Development — Build end-to-end retrieval-augmented generation systems: semantic chunking, query rewriting, multi-query retrieval, LLM re-ranking
- LLM Integration — OpenAI/OpenRouter APIs, LangChain, streaming, Pydantic structured outputs, cost-aware model selection
- Edge AI & Computer Vision — Real-time detection on constrained hardware (Jetson Nano, TensorRT)
Prompt Engineering | RAG | LLM APIs | Evaluation-Driven Development
- Designed 4 specialized prompts (chunking, rewriting, re-ranking, answer generation) with per-task temperature tuning and structured JSON output
- Achieved MRR 0.929, Accuracy 4.79/5 — a +24.8% improvement over a LangChain baseline
- Implemented cost-aware model selection, context grounding, streaming chat, and incremental document management
- Benchmarked with 150 test questions using MRR, nDCG, keyword coverage, and LLM-as-judge scoring
Repo: advanced-rag-chatbot | Demo Video
AI Classification | REST APIs | Workflow Automation
- End-to-end automation processing customer feedback with AI-powered sentiment analysis and categorization (OpenRouter)
- Integrated Google Apps Script, n8n, Telegram Bot via REST APIs and webhooks
- < 5 seconds processing time per submission with logging, retry logic, and status tracking
Repo: feedback-automation-ai
Computer Vision | TensorRT | Jetson Nano | MAVLink
- Two-stage cascaded detection (smoke → fire confirmation) to reduce false positives
- Jetson Nano deployment with TensorRT FP16; 10+ FPS inference
- Data fusion: detections + Pixhawk telemetry for geo-tagged alerts + autonomous LOITER
Repo: Real-Time-Fire-Smoke-Detection-Drone
Computer Vision | Robotics | MATLAB
- HSV segmentation + calibration (pixel → mm) for pick coordinates
- MATLAB forward/inverse kinematics, Arduino stepper control
- Includes Unknown rejection case to avoid misclassification
Repo: SCARA-Color-Sorting