← sources

Building a Fully Local AI Assistant That Doesn't Need a Wake Word - Part 2

source: reddit (reddit:developersIndia)published 2026-09-18 17:18ingested 2026-09-19 02:38

retained text

Hello Everyone,

Last time, I have create a design of my project  
In this part I have implemented the components, enhanced it using multiprocessing, multi-threading and pre-loading

First conversation, initial token latency is 12 to 13sec and from next conversation the latency is 8 to 10 sec

Yeah, this is still huge,What I loved most in this process is implementing the core concepts and seeing the decrease in latency

Working on this, My Idea/Imagination is kept getting refined from Autonomous AI Local Assistant to  
Goal and productivity Oriented assistant (user while on-boarding describes his/her goal and from there assistant will help/support user to achieve it)

What I am thinking now,  
Building the pipeline from scratch is taking time, So I will keep in background and I will try to implement this refined Imagination using FastRTC, Tauri, CrewAI, (try to integrate any existing agents)

Hearing about frequent layoffs has made me question whether I'm investing my time in the right direction. and or shell i consider GATE and other Govt Group exams. One of the reason for building this AI project is  preparing for a job switch into AI/backend/FullStack engineering

Please share, If you have any suggestions

Thanks a lot for your time

document fetched 2026-09-19 00:24 · canonical https://reddit.com/r/developersIndia/comments/1wjwca0/building_a_fully_local_ai_assistant_that_doesnt