<Yohanes Egi />
_
Building intelligent systems with LLMs, Computer Vision, and Multimodal AI.
Featured Projects
A selection of projects that demonstrate my skills in AI and software development.
Career Experience
My professional journey in the world of Artificial Intelligence.
Full Stack AI Engineer
July 2025 - Present
PT Verset Teknologi Nusantara
- Developed and deployed end-to-end AI systems covering backend services, LLM integration, and multimodal generative applications.
- Designed and implemented scalable AI pipelines from data ingestion to model inference, including locally hosted LLM deployment and optimization.
- Built intelligent AI agents for social media automation, enabling autonomous comment generation, content scheduling, and data scraping workflows.
- Created AI narrative tools capable of generating content from images, videos, text, and audio using multimodal models.
- Developed interactive AI chatbots with text and image generation capabilities, supporting real-time user interactions.
- Led the development of the Aiverse chatbot, a ChatGPT-like conversational AI powered by locally hosted LLMs with full control over data and inference.
- Architected modular and scalable AI services to support agent-based systems and production-ready AI products.
AI Researcher
August 2023 - June 2025
PT Ebdesk Teknologi.
- Joined the AI Research and Development team focusing on LLMs, multi-agent systems, and computer vision for AI product development.
- Designed multi-agent pipelines for large-scale data acquisition using Google, DuckDuckGo, and social media crawling tools.
- Built end-to-end RAG (Retrieval-Augmented Generation) workflows leveraging Qdrant and ElasticSearch to improve information retrieval accuracy and LLM response quality.
- Implemented LangChain-based agent systems to orchestrate data processing, reasoning, and automated analysis generation.
- Developed and fine-tuned multimodal AI models including Text-to-Speech (TTS) for natural voice synthesis, Text-to-Image for high-quality visual generation, and video mimic models for realistic avatar animation.
- Optimized large language models such as Qwen using parameter-efficient fine-tuning techniques (LoRA) and distributed training frameworks like DeepSpeed for scalable production deployment.
- Conducted data annotation and dataset preparation using Label Studio for computer vision tasks, including YOLO training and Vision-Language Model (VLM) fine-tuning.
- Collaborated on research-driven AI solutions, translating experimental models into production-ready systems.
Featured Hugging Face Spaces
Explore some of my interactive demos and models hosted on Hugging Face.
My Articles & Thoughts
A collection of my writings, tutorials, and thoughts on AI and technology.
Build a Modern RAG Pipeline in 2026: Docling + Qdrant Hybrid (BM25 + Dense) + AI Agent Step-by-Step Guide with Practical Code
Retrieval-Augmented Generation (RAG) continues to be the most practical way to build reliable, hallucination-resistant AI applications in 2026.
Single-GPU vLLM Deployment: Running Nemotron-3-Nano-30B on RTX A6000 An Architecture Deep Dive
NVIDIA’s Nemotron-3-Nano-30B-A3B (released December 2025) is a breakthrough in open-weight, efficient reasoning models. With a hybrid Mamba-Transformer + Mixture-of-Experts (MoE) architecture
My Tech Stack
My Tech Stack
A curated list of technologies I use to build modern, intelligent applications.