Skip to content
View avionicharshit-byte's full-sized avatar
19 followers · 161 following

Achievements

Achievement: Pull Sharkx2

Achievements

Achievement: Pull Sharkx2

Block or report avionicharshit-byte

Block user
See More

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. Markdown is supported. 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
avionicharshit-byte/README.md

Harshit

voice agents, speech models and the code around them

Python, C++, TypeScript, Go, Linux, Git

What I'm building

tinycue

Offline command understanding. No network, no model server.

A YAML file of example commands becomes a 250 KB model. It reads a sentence like "turn on the bedroom light" and answers set_light(room=bedroom, state=on) in 5 ms, or says unsure instead of guessing. English and Hinglish.

DocPilot

An AI document agent for loan officers, designed for Superleap CRM.

It asks the customer for documents over WhatsApp, reads them with OCR, fills the CRM, chases blurry or expired files, and only hands a case to the officer when a person has to decide. An interactive prototype with four real applicant scenarios.

openzl-agent

Format-aware compression, without the expertise.

Point it at your data. It writes the OpenZL description, trains a compressor, verifies the round trip byte for byte, and reports the savings against zstd and gzip.

contribution grid being eaten by a snake


Contributing to

pipecat-ai/pipecatOpen source framework for voice agents and realtime AI, by Daily
livekit/agentsA framework for building realtime voice AI agents
cactus-compute/needle14MB foundation model for phones, wearables, smart home and robots
snakers4/silero-vadPre-trained voice activity detector, the VAD most voice agents use
0xShug0/audio.cppPure C++ inference engine for audio models, powered by ggml
livekit/python-sdksLiveKit realtime SDKs for Python

also merged in: json · duckdb · diagram-design · fmt · cpp-httplib · growthbook · tailcat · Serial-Studio · openpanel · openzl


Voice AI

Voice agents: speech in, a reply out, fast enough for a phone call.
I fix bugs in the open source pieces they are built on, and build small models that understand commands offline.


Reach me

Pinned Loading

  1. tinycue tinycue Public

    Offline command understanding for tiny devices, under 1 MB

    C 15

  2. mcufit mcufit Public

    Check if an AI model fits on a microcontroller, before you flash it

    Python 2

  3. mcufit-bench mcufit-bench Public

    ground truth for mcufit: times tflite micro inference on real microcontrollers, per model and per layer

    C++

  4. openzl-agent openzl-agent Public

    point an agent at your data: it writes the OpenZL description, trains a compressor, and proves the savings against zstd and gzip

    Shell