mbachaud mbachaud

Building local-first AI infrastructure: Cymatix-context, ScoreRift, DriftWatch and BigEd. Less what it can do, More what if it could.

8 followers · 5 following

Highlights

Block or report mbachaud

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.

Close all issues, pull requests, and discussions opened by this user Content in all repositories owned by your account will be closed.

Add an optional note

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

Overview Repositories 14 Projects Packages Stars 10

Overview
Repositories
Projects
Packages
Stars

mbachaud/README.md

Michael Bachaud

I architect local-first AI systems and direct implementation agents through specifications, adversarial QA, instrumentation, and benchmark-gated iteration.

I don't hand a prompt to a model and ship what comes back. I run an adversarial engineering loop and hold the authority at every gate:

hypothesis → specification → implementation agents → instrumentation
  → benchmark → falsification → diagnosis → revised architecture

The receipts for that loop are in the issue trackers, not just the code: documented null results, isolated host variance, frozen benchmark versions, and issues I file against my own flagship features when the measurements falsify them.

What's here

Tier	Repo	What it is
🚩 Flagship	Cymatix Context	Local-first context/retrieval engine for LLM agents, no LLM call on the retrieval path. Shipped Python package; CLI + HTTP + MCP; ~3,400 tests; reproducible benchmark receipts.
🔧 Operational tooling	ScoreRift	Dual-layer audit: automated scoring + manual grading + reconciliation. The product isn't the score, it's watching the gap between them.
🏗️ Applied-architecture demos	DriftWatch, BookKeeper	Reference architectures — SOC 2 draft-generation assist, and a double-entry ledger core. Architecture samples, not certified/production products.
🧪 AI-development laboratory	BigEd	Explicitly not a product a lab for local-first multi-agent orchestration on ~11 GB VRAM edge hardware, file-based handoffs, HITL gates.
🔱 Forks & benchmark fixtures	headroom, AssetOpsBench, anker-solix, archestra-fork	Upstreams I build on or benchmark against. Labeled as forks, not pinned as my work.

How I split the work with models

Human (me): product thesis, constraints, architecture selection, acceptance criteria, experiment design, risk decisions, QA and falsification authority. Models: implementation, refactoring, draft documentation, test generation, code-review lenses.

I don't claim code authorship I don't have, and I don't donate the architecture authorship I do.

Pinned Loading

Cymatix-Context Cymatix-Context Public

Local-first context engine for LLM agents — retrieves, weighs, and compresses your codebase into the context window with no LLM call on the retrieval path. SQLite knowledge store; CLI + HTTP + MCP;…

Python 6 1
DriftWatch DriftWatch Public

SOC2 compliance helper - not a replacement to existing tools/certifications. Helps generate initial documents, and assists cross pillar compliance drift.

TypeScript 2
scorerift scorerift Public

Dual-layer audit system: automated scoring + manual grading + reconciliation

Python
BigEd BigEd Public

A reference architecture demonstrating local-first LLM orchestration for SOC compliance workflows. Built on ~11GB VRAM edge hardware using Ollama, featuring a multi-agent coordination pattern with …

Python 3

mbachaud (mbachaud) · GitHub