AI Engineer

Hao Hoang  • Peerlist

A hands-on cookbook with notebooks for fine-tuning open-source LLMs (Falcon-7B, FLAN-T5, Qwen) using PEFT (LoRA/QLoRA), SFT, and instruction tuning.
Covers training efficiency, evaluation (ROUGE metrics), and BitsAndBytes optimization.

LLM Finetuning Cookbook

Developed a collection of quantization techniques for open-source LLMs, achieving 75–85% model size reduction and 2–4× inference speedups while maintaining quality.
Covers GPTQ (4-bit), GGUF/GGML, ExLlamaV2, and 8-bit quantization.

LLM Quantization Zoo

Designed an agentic system using GraphRAG and fine-tuned LLMs for automated code reviews, reducing manual review time by 30%.
Built backend services with Kotlin and integrated AI features for context-grounded analysis using LangChain and Azure OpenAI.

Code Review Automation Platform (Approvia)

Created a real-time voice-interactive travel assistant powered by AssemblyAI, OpenAI, and ElevenLabs, capable of live speech recognition and natural voice responses.
Tech: Python, AssemblyAI, ElevenLabs, OpenAI, Streamlit

Realtime AI Travel Guide Voice Bot

Implemented a multimodal RAG system combining text and image embeddings to deliver rich contextual answers.
Deployed with Qdrant for efficient vector retrieval and DeepSeek-V2 for reasoning.

Multimodal RAG with DeepSeek & Qdrant

Built a medical imaging tool using MAIRA-2 multimodal AI to analyze chest X-rays and generate diagnostic reports automatically.

AI-Powered Chest X-Ray Report Generator

Developed an AI assistant that parses repositories, reads README files, and generates documentation automatically using LLMs and semantic retrieval.

RepoAgent: AI-Powered Documentation Assistant

Built a local ChatGPT-style chatbot using DeepSeek-R1 and Streamlit, with an animated “thinking” interface for transparent reasoning.