~/rs

Projects

flagship project

open-core / ai orchestration

SteerMesh

problem

Building reliable multi-agent AI systems is hard — agents drift, pipelines stall, and debugging distributed LLM workflows is near-impossible without visibility. Every role on the team (dev, BA, QA, ops, PM) needs agents, but no unified platform exists.

solution

An end-to-end AI agent ecosystem: define standards once in project.yaml, run them everywhere. 50+ specialized agents across 7 profiles, with a shared ACP protocol binding every interface to a single engine.

architecture

steer-runtime as the agent framework core, Kiro as the CLI runtime (ACP), Kite as the desktop companion (Tauri v2 + React 19 + Rust), Koda as the terminal TUI (Go + Bubbletea), Spool as the mobile PWA (React 19 + WebSocket), and Prompt Scorer as a zero-dependency Python service for real-time LLM quality evaluation.

ecosystem

steer-runtime

Core agent framework — 50+ agents, 7 profiles, spec-aware delegation, artifact pipeline with quality gates.

TypeScriptACPproject.yaml

Kiro

The underlying CLI agent runtime. All interfaces talk to Kiro via ACP — the shared engine behind every interaction.

CLIACP ProtocolJSON-RPC 2.0

Kite

Desktop companion — streaming markdown, live Mermaid diagrams, real-time prompt scoring, @agent multi-delegation, team workspaces.

Tauri v2React 19RustShiki

Koda

Terminal TUI — manage profiles, chat with agents, configure tokens, apply workspaces. Auto-update via LaunchAgent/cron.

GoBubbleteaLaunchAgent

Spool

Mobile PWA that turns your phone into a remote control for Kite. Scan a QR code and chat with Kiro from anywhere.

React 19PWAWebSocket

Prompt Scorer

Zero-dependency Python service — composite score (0–100) across 7 weighted dimensions with per-dimension reasoning and token estimation.

PythonZero deps7 dimensions

other projects

product · identity infrastructure

Verify SDK

Offline cryptographic validation of signed government ID barcodes (PDF417, ICAO VDS, Aadhaar Secure QR). DSA / X.509 signature checks — no network, no PII in the cloud. Ships as .dll / .so / .dylib / WASM with a YAML template engine for adding new document formats without code changes.

OfflineCryptographyDSA / X.509WASMYAML templatesLATAM IDs
brainbreaking.com/verify

product · entitlements infrastructure

Licensing-in-a-box

Hardware-bound, offline software licensing as a service. AES-256 encrypted, signed .lic files tied to MAC + hostname. Covers key issuance, secure distribution (signed time-limited download URLs), expiry, metering, and a dashboard — the same engine that meters and bills the Verify SDK.

Hardware bindingAES-256OfflineMeteringDashboard + API
brainbreaking.com/licensing

methodology · estimation framework

DRIFT

A 7-dimension estimation framework for LLM and agentic projects: Context complexity, Agentic depth, Corpus size, Orchestration, Retrieval depth, Validation rigor, Tool calls. Produces a complexity score and time formula — ground truth before you commit to a build.

LLM Estimation7 dimensionsAgenticMethodology
brainbreaking.com/#drift

research / open source

BrainWaves

Neural signal processing research — EEG pattern recognition using NeuroSky consumer devices. Built the signal acquisition pipeline and ML classification models for real-time brainwave analysis.

Neural Signal ProcessingPythonMLNeuroSky
BrainBreaking

collaboration

Conecta-Lab

Open collaboration platform for Latin American engineering communities. Contributing to tooling, documentation, and community infrastructure for regional tech ecosystems.

Open SourceDistributed SystemsCollaboration
Conecta-Lab