Home / Products / Corporate Bullshit Agentic Generator
← Back to PEACH STUDIO Runs on NVIDIA DGX Spark · Dell GB10

Corporate Bullshit
Agentic Generator

The Corporate Bullshit Generator is back — and it's AI now. Pick a topic and a persona. Receive confident, scalable, mission-critical nonsense — written by a local LLM, spoken aloud, and lip-synced to video. 100% on-device, no cloud.

View on GitHub How it works
CBAG generating a visionary-CEO monologue about AI strategy, entirely on a local DGX Spark
CBAG controls: topic, persona, length, voice, and output options including clone my voice and upload a portrait
Topic × persona × voice — your nonsense, your way
CBAG voice-cloning dialog asking you to read a short sample aloud
Clone your own voice from a short sample

What it does

Language

Local LLM monologues

An open-weight model running on-device writes confident, scalable, end-to-end synergy — in short, medium, or long. Buzzword density: maximum.

Personas

Persona engine

Visionary CEO, McKinsey consultant, and friends. Each persona has its own voice, cadence, and special brand of mission-critical nonsense.

Speech

Spoken aloud

Local text-to-speech reads the monologue back to you. Pick a default voice — or have it read in your own.

Voice clone

Clone any voice

Read one short sentence aloud and CBAG clones the voice locally — the same capability an attacker would use for vishing and impersonation, shown end-to-end.

Synthetic video

Talking-head deepfake

Add a front-facing portrait and CBAG lip-syncs the audio into a talking-head clip — a controlled demonstration of how convincing local deepfake video has become.

Edge

100% on-device

LLM, speech, voice clone, and video all run locally on a NVIDIA DGX Spark or Dell GB10. Open models, no cloud, nothing leaves the box.

Why it matters

A fun toy — and a serious offensive-security demo.

Behind the jokes, CBAG is a deliberate demonstration. The same pipeline that writes a buzzword monologue proves that convincing voice cloning and deepfake video can now be produced end-to-end on a single local machine — a NVIDIA DGX Spark or Dell GB10 — with no cloud, no third-party API, and nothing ever leaving the device. If a red team can stand this up on commodity AI hardware in an afternoon, so can an adversary. Understanding that threat is the first step to defending against it.

Controlled

Built and run as a closed research environment — not a public service. No hosted generation, no uploads to anyone else's servers.

Local & air-gappable

Everything executes on the Spark / GB10 itself, so synthetic-media capability can be studied entirely offline.

Threat-informed

A hands-on look at vishing, impersonation, and synthetic-media social engineering — the attacks defenders now have to anticipate.

How it works

1
Pick a topic and a persona. Type a theme — "AI strategy", "working together" — choose a persona, length, and voice.
2
A local LLM writes the monologue. Buzzwords are generated, then refined into a polished, straight-faced corporate monologue.
3
Local TTS speaks it. The text is read aloud on-device — in a default voice, or in a clone of your own.
4
Optionally, it becomes a face. Add a portrait and the audio is lip-synced into a talking-head video. Skip voice or video any time.
5
All of it stays local. Every stage runs on a DGX Spark or Dell GB10 with open models. No API keys, no cloud, no data leaving the machine.

Runs on

Hardware
DGX Spark / Dell GB10
Models
Open weights, local
Pipeline
LLM · TTS · Video
Cloud
None

Get the code

CBAG is an open-source demo of a fully local, agentic AI pipeline — LLM, speech, and talking-head video — built for the NVIDIA DGX Spark and Dell GB10. Submitted to the NVIDIA DGX community playbooks.

View on GitHub

Built by PEACH STUDIO · runs 100% on-device, no cloud