Skip to content
View andrei-tolkachev's full-sized avatar
1 follower · 0 following

Block or report andrei-tolkachev

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
andrei-tolkachev/README.md

Andrei Tolkachev

Building LoraAI · Local-first multimodal AI

C++ / GGML / CUDA · Voice & vision · AI agents · Video production

I build practical AI software that runs on your own hardware, with a focus on native model inference and creative workflows.

Website · Русская версия

Featured project · LoraAI

LoraAI is my independently developed, local-first AI platform. Its goal is to bring multimodal understanding, voice interaction, tools, memory, and production workflows together in one system — without making cloud inference a requirement.

Area Focus
Multimodal inference Local language, audio and vision models, including native Qwen3-Omni integration with streaming speech.
Voice & video interaction Real-time-feeling conversation, streaming text/audio, and video-call integration. Audio conversation and video-call operation have been tested in my LoraAI environment.
Apps & agents Desktop/mobile experiences, tool integrations and a growing local agent workflow.
Creative production AI-assisted video editing, DaVinci Resolve automation and production-oriented tools.
Memory & control Context continuity, controlled runtimes and private infrastructure — areas of ongoing development and validation.

Status: LoraAI is actively developed; components are at different stages of maturity. The complete platform, private services and production infrastructure are not open-sourced here. Selected independent modules and developer tools are published below.

Learn more about my work

Public projects

Project What it does
Qwen3-Omni Native for llama.cpp Unofficial native C++/GGML Talker → Code Predictor → Code2Wav speech integration for a pinned llama.cpp revision. First release v0.1.0.
VoiceBridge Local desktop push-to-talk dictation built around Qwen3-ASR.
AvtoCvetocor Offline HDR → Rec.709 normalization and basic automated color correction for video.
MaskGen Experimental video masks, foreground extraction and background replacement.

Licensing differs by repository. Public availability does not necessarily mean the source is open for redistribution or modification.

How I work

Local-first where practical · evidence-backed testing · reproducible builds · transparent limitations · respect for upstream licenses and authorship.

Contact & project information: videomontazher.com

Independent developer and community projects. Qwen3-Omni Native is not an official release of QwenLM or ggml-org.

Popular repositories Loading

  1. MaskGen_VideoMontazher MaskGen_VideoMontazher Public

    Tool for object extraction from video. Supports output as B/W mask, object on green background, object with alpha channel, and background replacement. Works via GUI.

    Python 1

  2. AvtoCvetocor AvtoCvetocor Public

  3. VoiceBridge VoiceBridge Public

    Offline voice dictation for Ubuntu & Windows powered by Qwen3-ASR · Локальный голосовой ввод

    Python

  4. andrei-tolkachev andrei-tolkachev Public

  5. qwen3-omni-llama-cpp-native qwen3-omni-llama-cpp-native Public

    Forked from ggml-org/llama.cpp

    Independent native Qwen3-Omni speech and True Full Duplex for llama.cpp: concurrent Thinker/Talker, GPU streaming audio, playback-confirmed history and standalone upgrade kit.

    C++

  6. llama-cpp-moe llama-cpp-moe Public

    GGML-compatible MoE optimizations for llama.cpp: adaptive prefill/decode memory budgets, CPU kernels, reproducible measurements and build tools.

    Python