Los Angeles · est. 2024
We build products around two convictions: that the unit of value in our era is a decision, not a list — and that every product now has two readers, a human and an agent.
00 / Two bets
Bet 01 · Decision as a service
The unit of value is a decision, not a list.
For the last decade, service discovery looked like a list: twenty plumbers on Yelp, fifty proposals on Upwork, a feed of related results on Amazon. The decision was always handed back to the customer.
AI changes that. Picking the right two is itself an AI decision — calibrated to intent, urgency, budget, and trust. We design products that ship a decision, not a search result.
Bet 02 · Two readers per product
Every product now has a human reader and an agent reader.
For the last decade, software was written for a single audience: a person on the other end of a screen. AI changes that too.
Every consequential product now has an agent reader as well — a model that has to parse it, trust it, and act on it on behalf of someone. We design the schema, the API, and the consumer surface in lockstep. Same product, two readers, day one.
Products
01 / Security
Hardening the rails that AI agents run on. Static scanners, runtime guards, and a rating system for the tools, skills, and servers an autonomous agent is allowed to touch.
Scan, rate, and harden MCP servers for AI agent safety.
Static scanner with 46 security codes, runtime tool-call firewall, transparent MCP proxy, PII / secret DLP, structured audit trail, and an allowlist-mode agent security checker. Published to PyPI.
Security index and certification for AI agents, tools, and skills.
Rating engine (F → A) with calibrated ground truth, enterprise dashboard for multi-agent fleet visibility, certification API for badge issuance, behavioral baselines and anomaly detection. EU AI Act-ready reporting.
Curated list of MCP server security resources, tools, and research.
02 / Martech
Marketing, media, and matching — rebuilt for an era where most discovery starts in a model and most fulfillment is mediated by an agent. We run the brand-monitoring, the affiliate network, the bilingual marketplace, and the consumer-facing surfaces in one stack.
AI-Native Martech Platform — AI Brand Monitor + 9,800+ provider directory.
Helps brands win in the age of AI recommendations. Measures how an LLM talks about you, identifies query gaps, and runs the underlying local-business intelligence database powering several downstream products.
An AI-native CPS affiliate network for North America. Flat 20% network fee, USD settlement, bilingual operations.
Connects performance-marketing brands directly to publishers across newsletters, communities, social creators, and service professionals. Direct-brand only — no sub-affiliate layer.
A bilingual freelance marketplace bridging Mandarin-speaking talent and global brands.
AI-native curated marketplace with managed services (Stripe Connect escrow, Wise payouts, 1099). Dual cohort design: North American premium expertise and a global commodity skill pool. Agent-to-agent matching from day one.
A curated remote-work aggregator for Mandarin-speaking talent. Filtered, translated, classified.
An AI-matched local services marketplace. Property-owner intent routed to vetted, licensed pros.
A bilingual life assistant for the overseas Mandarin-speaking community — visa, DMV, banking, daily living. Plus a personal twin agent.
Bilingual deals editorial. Verified price drops only — no fabricated discounts.
03 / Finance
Autonomous research and execution stacks for capital markets. Overnight quant labs that discover alpha factors, an event-driven explainability shell that turns price moves into auditable causal chains, and adjacent prediction systems.
AI hedge fund — autonomous trading research and execution.
Private fund. Not soliciting investors. Listed here for portfolio completeness.
Autonomous alpha factor discovery via overnight research loops.
Long-running agent research process: hypothesis generation, factor mining, backtesting, and report synthesis run continuously without human prompting.
Explainable alpha & risk — event-driven drawdown / gainer attribution with event-probability narratives.
Turns 'why did this move?' into a causal chain a human (or an LLM auditor) can verify.
AI sports prediction — model-driven edge discovery across global leagues.
04 / Local
Boring services, AI-era distribution. We operate a small fleet of single-niche, single-city service sites — built from a shared template, ranked organically, and rented to operators — and a Chinese-language directory for overseas life essentials.
Single-niche local service sites — pool cleaning, drain cleaning, AI-era SEO standards built in.
Properties: Chino Hills Pool Cleaning, Irvine Pool Pro, Pasadena Drain Cleaning, Glendora Pool Cleaning, Rancho Cucamonga Pool Cleaning.
A Mandarin-language local-life services directory for the overseas community.
Open source
The pieces of our work that are useful to the broader AI-safety and developer communities live on GitHub and PyPI.
About
We build, ship, and operate the products listed above ourselves — no outside investors, no growth-at-all-costs roadmap.
We work across the full stack: model evaluation, agent tooling, data infrastructure, and consumer-facing product. Some of our work is open source. Some is operated as private services. Some is internal infrastructure that other parts of the studio depend on.
