Steven Pajewski

Steven Pajewski

you can call me Vel

Curated project index

A compact map of the systems I build: agent workbenches, local-first AI, simulations, MCP tooling, authority layers, governance scaffolds, and trading control planes. Each entry names its status and the proof boundary it is meant to demonstrate.

AllAgent workbenchesLocal-first AI / MLSimulation systemsMCP / toolingAuthority / governanceTrading / control planes

All proof lanes

Showing 9 of 9 projects. Read from the public indexes — revalidated on a daily window. Index check value eddf6d2b.

Local-first AI / MLHackathon build

CareSight

Local-first caregiver awareness prototype for home observations, human review, journaling, and bounded handoff paths.

What it proves

Sensitive care events can be modeled as local, inspectable records before any external alert or handoff is approved.

  • Python
  • MLX
  • SQLite
  • local vision
  • human review
Repo
Local-first AI / MLPrototype

yolo26-mlx-swift

Swift port work around YOLO26 MLX for Apple-platform local vision experiments.

What it proves

Local model capability can move closer to the app runtime instead of being treated as a remote service by default.

  • Swift
  • MLX
  • local vision
  • Apple platforms
Repo
MCP / toolingActive

vel-mcp

Modular MCP tooling scaffold for local or cloud-backed senses: vision/OCR/grounding, control, brain, speech, and privacy gateway layers.

What it proves

Agent tools can be split by trust boundary so each capability has its own permission surface, worker lifecycle, artifacts, and audit trail.

  • TypeScript
  • MCP
  • local workers
  • vision evals
  • audit logs
Repo
Authority / governanceHackathon build

ClearIntent

Authority layer for autonomous agents that turns intents into scoped, signable, auditable actions on EVM chains.

What it proves

Agent execution can be wrapped in intent schemas, policy checks, identity, signing surfaces, provider evidence, and explicit gaps.

  • TypeScript
  • EVM
  • EIP-712
  • ENS
  • 0G
  • KeeperHub
Repo
Authority / governanceActive

thc-methodology

Project-agnostic reliability methodology for surfacing hidden trust in software systems, docs, and agentic workflows.

What it proves

Truth, Hardening, and Clarity can become a reusable review language for source-of-truth boundaries, claims, and operator readiness.

  • docs
  • templates
  • scorecards
  • local audit skills
Repo
Authority / governanceActive

project-scaffold

Reusable governance scaffold for fast-moving open-source, hackathon, and AI-assisted projects.

What it proves

A repo can start with contracts, decisions, audits, quality gates, and agent instructions before implementation pressure blurs authority.

  • TypeScript
  • contracts
  • templates
  • quality gates
Repo
Agent workbenchesActive

CommandCode-UI

Native-feeling desktop workbench for a terminal-first coding agent, with PTY health, sessions, permission state, and receipts surfaced in the UI.

What it proves

A CLI agent can remain the execution source of truth while the operator gets a calmer cockpit for setup, runtime state, and diagnostics.

  • Electron
  • React
  • TypeScript
  • PTY
  • local receipts
Repo
Listed in an index but not shown (2)
  • Agent orchestration workbench — confirmed absent — excluded, recorded for review
  • Vel-Labs/civulacrum — confirmed absent — excluded, recorded for review