Building a Personal AI Companion at 75: A Hands-On Journey into Local AI with RAG and a Talking Avatar
In an age where technological advancements often seem reserved for the tech-savvy, one retiree’s experience proves that age and coding experience are no barriers to creating sophisticated AI tools. At 75 years old, I embarked on an exciting project to build my own private, local AI assistant—complete with a talking avatar and a question-answering system tied to my personal knowledge base. This is a reflection on my journey, the final setup, and encouragement for others to explore AI regardless of their background.
My Motivation and Background
A few weeks ago, I knew virtually nothing about GitHub, command-line interfaces, or local large language models (LLMs). I’m not a coder by profession; my interests are more about fun and enhancing my hobbies—specifically, creating an AI companion for daily interaction and a specialized bot for my longstanding game wiki.
Overcoming Initial Challenges
Admittedly, the learning curve was steep. The process involved numerous errors, extensive reading, and reaching out for help from various AI assistants—including my go-to, DeepSeek AI. But perseverance paid off. Here’s what I managed to establish:
My Final AI Setup
-
Local Language Model (LLM) Environment:
Running the LM Studio platform with a 14-billion or 32-billion parameter DeepSeek model on my NVIDIA RTX 4090 graphics card. Impressively, this setup is completely offline, ensuring privacy and control. -
Interactive Avatar with Voice:
Implementing a Live2D avatar named Mao, which not only looks engaging but also speaks, making interactions more personable and fun. -
Wiki Bot Integration:
Utilizing AnythingLLM alongside Ollama to develop a separate web-based bot that ingests content from my gaming wiki through a simple browser extension, enabling one-click data feeding. -
Retrieval-Augmented Generation (RAG):
Configuring the system to answer questions based on my personal documents, with citations included for reference—making the AI not just responsive but also trustworthy.
What I’ve Accomplished
Through this process, I can now proudly say that I:
- Successfully built and ran a local LLM tailored to my needs
- Configured GPU acceleration features like CUDA and VRAM offloading for optimal performance
- Implemented RAG with document embedding for accurate, context-aware responses
- Integrated a browser extension for seamless wiki data ingestion
- Trained the AI on my own knowledge base—an ongoing journey!
- Debugged various components including Python scripts, YAML configurations, WebSocket connections, and API integrations
A Message to Fellow Enthusiasts
While these tasks might seem straightforward for experienced programmers, for a newcomer like myself, it was a milestone filled with learning and persistence. Thanks to the support of helpful tools like DeepSeek AI, I was able to achieve a highly functional, personalized AI environment.
Inspiration for Others
If a 75-year-old retiree with no coding background can accomplish this, it opens the door for anyone interested in exploring AI. Technology is for everyone, and with patience, curiosity, and the right resources, you can create something truly remarkable.
Stay Updated
Interested in following my ongoing journey? Visit my blog for real-time updates and deeper insights into my AI adventures: The Blog.
Conclusion
My experience demonstrates that age and technical know-how don’t have to hold you back. With determination and a little help, building your own local AI is within reach. So why not start today? The future is yours to shape.
