🤓 Latest Submissions
HomeStar — CRE Rental Intelligence
HomeStar is a commercial real-estate rental-market intelligence platform built around a validated insight: demand leads rent. Job postings, JOLTS quit-rates, WARN layoff notices, and employment turn 5–7 months before rent growth does, and HomeStar's engine detects those turns. A walk-forward backtest achieves a 79.3% hit rate across 17 markets (46 of 58 confirmed turns), and the result held byte-identical when the sample doubled from 9 to 17 metros — evidence the signal is real, not overfit. The product has three layers: full market analysis across 17 metros and six demand signals; an agentic drill-down where you ask an analyst agent why any market is firming or softening and every quantitative answer comes from a validated tool with sources, never the model's invention; and backtesting that measures signal strength rather than asserting it. The discipline is the differentiator. We tested every public leading indicator against pre-stated criteria and dropped what failed — permits (pro-cyclical), wages (wrong-signed in 3 of 9 markets), rent-vs-own (mechanical correlation). The backtest also taught us the engine is an early-warning tool for avoid/build decisions, not a trading signal. Bright Data Web Unlocker powers the live layer: LinkedIn job postings and apartments.com concessions scraped from anti-bot sources — a live public-data proxy for signals institutions license from CoStar and RealPage. The result that matters is the convergence: the Sun Belt markets the engine flagged on labor fundamentals (Atlanta 90%, Phoenix 88%, Austin 83% concessions) are the same ones running hot, while constrained coastal markets run low (San Francisco 28%, New York 24%, Boise 15%). Two independent data sources, two methods, the same answer.
31 May 2026
Gyasss — Pay-per-query oracle on Circle + Arc
Americans spend $560B per year on gasoline. SF drivers leave $200–$400 annually on the table by not routing to the cheapest nearby station. The dominant data source — GasBuddy — relies on self-reported submissions with no economic incentive for accuracy. Most prices are stale or wrong. Gyasss is a transaction-verified gas price oracle that fixes this with two parallel economic loops, both running on Circle Nanopayments and Arc: INBOUND: Users submit price reports and earn USDC cashback by data freshness — $0.005 for redundant data up to $0.50 for breaking a 24-hour stale-station bounty. A consensus-weighted oracle combines the last 30 reports per station with time-decay weighting, so no single user can move the price. High-value bounty reports require a $0.10 USDC stake; outliers get slashed. OUTBOUND: Two oracle endpoints — cheapest-gas and cheapest-parking — serve real-time queries to AI agents via x402. Each query costs $0.001 USDC, settled in batches via Circle Gateway. Any developer worldwide can query our oracle now using AIsa's open-source x402 client. The bidirectional architecture is the key insight: same Nanopayments primitive, opposite roles. We are simultaneously an x402 seller AND a USDC sender. To our knowledge, no other entry uses Nanopayments in both directions. Parking is a live second vertical proving the platform thesis. Roadmap: any commodity with dynamic pricing where transaction-verified data beats self-reporting. The recommendation engine is route-aware via Mapbox Directions, computing real driving detour distance and time, then netting against value-of-time and detour gas cost. WHY THIS FAILS WITH TRADITIONAL GAS: We pay users $0.005–$0.50 per data point and charge agents $0.001 per query. On Ethereum ($2–$5/tx), both sides are 1000x underwater. On L2s ($0.01–$0.05), the agent market doesn't exist and cashback margin is negative. Only Arc's batched settlement makes both loops positive-margin.
26 Apr 2026
