Projects
Using AI to strengthen the editorial process, not replace editorial judgment.
Editorial judgment. Using AI to surface context, compare source material, and prepare more deeply, while the calls stay human.
Verification culture. Provenance, citations, and source quality first, with clear limits on what agents can do in financial media.
Local-first tools. Private systems for memory, research, and operational context, where sensitive data stays close to the user.
AI as an operating layer.
I’m interested in AI systems that work across real surfaces: inboxes, calendars, repositories, research, notes, editorial pipelines, and private data. Useful AI should preserve context, cite its sources, surface the next action, and keep humans in control of important decisions.
+-------------------+ +---------------------------+ +-----------+ +--------------------+ | PRIVATE SOURCES | | RUNS LOCALLY | | APPROVED | | CONTROLLED OUTPUTS | | Notes | --> | Memory + task state | --> | review | --> | Publishing | | Inbox | | Source provenance | | intent | | Payments | | Repositories | | Local inference + agents | | limits | | Sync | | Research | | | | | | Open networks | +-------------------+ +---------------------------+ +-----------+ +--------------------+
Private sources
- Notes: Notes, drafts, inbox context, calendar commitments, repositories, research, and documents stay close to the user.
- Inbox: Inbox, calendar, and notes become working surfaces instead of isolated tabs.
- Repositories: Code and research connect without turning private work into a cloud-first dataset.
- Research: Source material travels with the work, so provenance is visible when decisions are made.
Runs locally
- Memory + task state: Memory and task state turn scattered private context into a working layer that persists across tools.
- Source provenance: Source provenance keeps claims attached to evidence trails, citations, and original material.
- Local inference + agents: Local inference and agent workflows reason over context before anything leaves the machine.
Approved
- review · intent · limits: Human approval is the boundary for consequential actions: review, intent, and limits before anything goes out.
Controlled outputs
- Publishing: Publishing stays a controlled output, never an automatic consequence of generated text.
- Payments: Payment and wallet actions sit behind explicit user intent and approval.
- Sync: Sync moves selected state between surfaces without surrendering the whole private workspace.
- Open networks: Open networks matter when tools should survive outside one platform or intermediary.
The Coin Flip
Heads: AI fixes the world. Tails: it doesn’t matter. Either way, you deserve better questions than your feed is asking.