Activity
-
Slack status: Working from China 🇨🇳 Next week, most of our hardware team will be on the ground in China. MEs, TPMs, our CPO - each person with a…
Slack status: Working from China 🇨🇳 Next week, most of our hardware team will be on the ground in China. MEs, TPMs, our CPO - each person with a…
Liked by Caglar O.
-
Big news: I’ve officially joined the recruiting team at Peregrine! 🚀 At Peregrine, we aren't just building another software platform; we’re…
Big news: I’ve officially joined the recruiting team at Peregrine! 🚀 At Peregrine, we aren't just building another software platform; we’re…
Liked by Caglar O.
-
We're growing quickly and hiring like crazy at Cyclops with 20+ people already on the team. We have remote roles across compliance, marketing…
We're growing quickly and hiring like crazy at Cyclops with 20+ people already on the team. We have remote roles across compliance, marketing…
Liked by Caglar O.
Experience
More activity by Caglar
-
Lights out 🚀 We’ve got some BIG news. Eight Sleep joins forces with the Aston Martin F1 Team The strongest partnerships happen when two teams…
Lights out 🚀 We’ve got some BIG news. Eight Sleep joins forces with the Aston Martin F1 Team The strongest partnerships happen when two teams…
Liked by Caglar O.
Other similar profiles
Explore more posts
-
Ron Pragides
"Shoppers are strategically shifting toward more gift card purchases to maintain holiday gifting traditions amid economic pressures, according to new research from Blackhawk Network (BHN), a leading global provider of branded payment solutions. BHN's annual holiday forecast, which surveyed over 2,100 U.S. consumers, reveals that while 13% are purchasing gifts for fewer people due to financial constraints, overall holiday gift spending continues to rise and gift cards will represent 39% of total holiday budgets—a 12% increase over last year." https://lnkd.in/gQxrpy35
-
Jay Garmon
This a fascinating and innovative use of transformer architecture, but I'm really curious how much the false-positive rate shifted using LMMs. That's a glaring omission from the write-up. We've all been annoyed when our card issuer blocked a legitimate transaction when it "looked" like fraud. Managing fraud algorithms is incredibly hard because you absolutely prefer false positives to false negatives (annoyance is better than theft), but you can't let false positives creep above a tolerable threshold without ruining the experience of legitimate customers. This write-up declares the LLM-like fraud monitor far more capable of nailing true positives (especially for large-scale "testing attacks" wherein fraudsters test whether stolen card numbers are still active), but neglects to mention if it saw an intolerable spike in false positives to go along with it. Given that LLMs/transformers are infamous for hallucinations, "imagining" fraud would seem an outsize risk for these systems. That they didn't address that at all is curious at best.
-
Michele Wallace
The latest credit card bill in congress is causing a stir. I'll share some thoughts based on my background in payments. 𝐂𝐚𝐫𝐝 𝐍𝐞𝐭𝐰𝐨𝐫𝐤 𝐚𝐧𝐝 𝐂𝐚𝐫𝐝 𝐈𝐬𝐬𝐮𝐞𝐫 In the U.S., there are four primary credit card networks: Visa, Mastercard, Discover, and American Express. Discover and American Express operate as both the network and the card issuer, while Visa and Mastercard rely on financial institutions to issue cards. Among issuers, Chase holds the largest card portfolio, bolstered recently by its acquisition of the Amazon card program. 𝐓𝐫𝐚𝐧𝐬𝐚𝐜𝐭𝐢𝐨𝐧 𝐀𝐮𝐭𝐡𝐨𝐫𝐢𝐳𝐚𝐭𝐢𝐨𝐧 𝐚𝐧𝐝 𝐏𝐫𝐨𝐜𝐞𝐬𝐬𝐢𝐧𝐠 Large issuers such as Chase both authorize and process transactions for the cards they issue. Smaller issuers often partner with larger financial institutions or third-party processors to handle authorization and transaction processing on their behalf. 𝐌𝐞𝐫𝐜𝐡𝐚𝐧𝐭 𝐅𝐞𝐞𝐬 Merchants pay transaction fees on card payments that consist of three primary components: ✅Interchange fees paid to the card issuer ✅Network fees paid to the card network such as Visa, Mastercard ✅Processor markups paid to companies such as Square, Stripe, or Chase One insight that often surprises people: when a consumer chooses “credit” rather than “debit” when using a debit card, the card issuer typically earns more on the transaction. Should that distinction really impact the fee charged to merchants? 𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭 𝐑𝐚𝐭𝐞 Interest rates on unpaid balances are influenced by several factors. Credit score is the most obvious, but rewards programs also play a role with richer rewards often correlating with higher interest rates. 𝐌𝐲 𝐏𝐞𝐫𝐬𝐩𝐞𝐜𝐭𝐢𝐯𝐞 Card issuers exert significant control over both merchant costs and consumer interest rates. If the goal of legislation is to reduce costs for merchants, perhaps greater focus should be placed on card issuers and processors rather than card networks alone. Additionally, while credit cards represent unsecured debt, there should be reasonable limits on interest rates. With the prime rate around 6.75%, it raises the question of whether interest rates ranging from 19% to 36% are truly necessary. I pay my balance in full each month, so interest rates do not directly affect me. However, I don’t believe consumers who carry balances should be disproportionately penalized. A return to simpler, no-reward cards may not be such a bad outcome if it leads to fairer pricing for both merchants and cardholders. https://lnkd.in/gpjbUpqa
1 Comment -
George Gilbert
I stopped by the NYC Founders Club meetup last night to chat with AI founders about monetization. Lots of folks wanted to understand what is the "winning" strategy to balance acquisition, growth, and healthy margins. Stripe interviewed 2k businesses and we are sharing out the results today. Our data shows that AI companies adopt hybrid pricing models (subscription plus usage) and change their prices ~2x more frequently than others. For the full set of trends, check out our report: https://lnkd.in/eReqAfkG
6 Comments -
Rhonda Coleman Albazie
Below are AI-driven ways to build better, more flexible housing value prediction systems Part I that go beyond simple baseline models. View these two types of approaches: • Market prediction markets (like Polymarket) which reflect speculative betting on future values. • Real-estate platforms (like Realtor.com, Zillow) that use data and AI to generate forecast models and valuation estimates. 🧠 1) Build a Machine Learning Forecasting Engine Instead of relying on single indices: ✅ Collect Rich, Diverse Data Good predictive models need many signals — not just comps. Include: • Historical sale prices & transaction volume • Macro economic indicators (mortgage rates, CPI, unemployment) • Supply metrics (new listings, inventory changes) • Mortgage origination/approval trends • Demographics & migration • Local city/town data (schools, crime, traffic, walkability) • Spatial / neighborhood context (amenities, transit, zoning) The more structured and temporal the data, the better your models can learn long-term patterns. ⸻ 📈 Use Strong Machine Learning Models Traditional regression is weak for real estate’s highly nonlinear patterns. Instead use: • Gradient Boosted Trees (XGBoost, LightGBM) — widely shown to outperform linear baselines. • Random Forests (good for handling mixed data types) • Neural Networks / Deep Learning (good for non-linear trends, especially across multiple features) • Spatial ML / Geo-models incorporating geographic context (e.g., graph networks). These models can capture complex interactions that simple indices or linear trend models miss. ⸻ 📊 2) Feature Engineering & Alternative Data AI beats static models when you engineer features that represent real market forces: 📌 Examples of Predictive Features Feature Type Why It Matters Month-over-Month price % Shows velocity of change Mortgage rate spreads Influences borrowing demand Home construction permits Future supply indicator School ratings & walkability Drives desirability Social sentiment (online / news) Early indicator of demand shifts Nearby transactions in micro-radius Local power law effects These go beyond listings and comps — giving your model foresight not just hindsight. ⸻ 🌐 3) Leverage APIs & Real-Time Data Feeds Use APIs from: • Zillow API (Zestimate, comps, rental data) • Mashvisor / AirDNA (for detailed STR and rental trends) • Census, MLS, county tax assessor info • Economic data sources (Federal Reserve, St. Louis Fed FRED) Streaming fresh data helps your model adapt quickly — unlike prediction markets which update after expectations shift.
-
Anthony Zak.
Solace Health just raised $130M in Series C funding, reaching a $1B valuation. Founded by Jeremy Gurewitz 👨💼 after his family struggled to navigate his mother’s cancer care, the company connects Medicare patients with trained healthcare advocates who help coordinate care, manage insurance appeals, attend appointments, reduce medical bills, and handle the complexity most patients are left to figure out alone. Solace now supports 20,000+ patients per month, with 98% reporting better outcomes. Most patients don’t pay out of pocket. The new funding will expand its national advocate network and deepen partnerships with payers and providers to embed advocacy earlier in the care journey. The broader shift is clear: patient advocacy is moving from a “nice-to-have” service to a core layer of healthcare infrastructure. https://lnkd.in/dJVZ3n_t
6 Comments -
Thomas D'Eletto
The recent article from Marty Cagan on Google's product model is excellent (as always): small, empowered teams own outcomes—not features—via continuous discovery, strong vision, and rapid delivery. True from crypto wallets as well and very important for designing products that meet the needs of wider audiences. https://lnkd.in/egAS3hg5
-
Andriyan Bobryk
DeepSeek’s rise has been wild to watch. From hedge fund roots to topping app store charts, all while running models cheaper and faster than the big names? That’s no small feat. Their focus on reasoning models is especially interesting and definitely shaking things up in the AI space. #AI #DeepLearning #Innovation #TechTrends #MachineLearning #Startups https://lnkd.in/dDfmM99D
-
Bharat Kumar Kondapalli
𝐖𝐡𝐞𝐧 𝐲𝐨𝐮𝐫 𝐛𝐮𝐫𝐫𝐢𝐭𝐨 𝐛𝐨𝐰𝐥 𝐤𝐧𝐨𝐰𝐬 𝐲𝐨𝐮𝐫 𝐭𝐚𝐱 𝐛𝐫𝐚𝐜𝐤𝐞𝐭 I was reading summary of Chipotle’s latest earnings report released about 2 weeks ago and one statement from Chipotle's CEO stopped me: 60% of their core customers earn over $100,000 a year. 𝐌𝐲 𝐟𝐢𝐫𝐬𝐭 𝐫𝐞𝐚𝐜𝐭𝐢𝐨𝐧 𝐰𝐚𝐬 𝐩𝐮𝐫𝐞𝐥𝐲 𝐚𝐬 𝐚 𝐜𝐮𝐫𝐢𝐨𝐮𝐬 𝐜𝐮𝐬𝐭𝐨𝐦𝐞𝐫:How do they even know that? I don’t remember filling out a financial profile to get a side of guac. I just tap my card, grab my bowl, and walk out. But then, the engineer in me took over. I started thinking about...now, really- 𝐇𝐨𝐰? from a systems architecture and data-aggregation perspective. It hit me, The Architecture of a 𝐒𝐡𝐚𝐝𝐨𝐰 𝐏𝐫𝐨𝐟𝐢𝐥𝐞 that Chipotle isn't just a restaurant. In 2026, it’s a sophisticated data aggregator as well. Here is how that 60% figure likely moves through their pipeline: 𝙏𝙝𝙚 𝙩𝙧𝙖𝙣𝙨𝙖𝙘𝙩𝙞𝙤𝙣𝙖𝙡 𝙃𝙖𝙨𝙝: When you swipe your card, you’re generating a unique, anonymized identifier for them. 𝙏𝙝𝙞𝙧𝙙 𝙋𝙖𝙧𝙩𝙮 𝙀𝙣𝙧𝙞𝙘𝙝𝙢𝙚𝙣𝙩: Chipotle (and most major retailers) use 𝐃𝐚𝐭𝐚 𝐄𝐧𝐫𝐢𝐜𝐡𝐦𝐞𝐧𝐭 services from firms like Acxiom, Experian, or Cardlytics. They take that anonymized card ID and 𝐣𝐨𝐢𝐧 it with massive, external datasets including census data, property values, and credit-level spending patterns. 𝙏𝙝𝙚 𝙆-𝙎𝙝𝙖𝙥𝙚𝙙 𝙇𝙤𝙜𝙞𝙘: You must have already heard about K-shaped economy. By mapping where that card is used (high-end grocers, luxury retail and other restaurants) and the neighborhood of the cardholder (that's easy - based on the proximity you most shop at. That gives them most likely zip code you live in and average house value or rent prices there), they can statistically estimate household income with incredible accuracy without ever asking you directly about your income. Why this matters? From 𝐂𝐮𝐬𝐭𝐨𝐦𝐞𝐫𝐬 to 𝐔𝐬𝐞𝐫𝐬 During the earnings call, CEO repeatedly referred to us not as customers, but as core users. He already got a lot of slack on media for that. As an engineer, that terminology shift is telling. It implies we are part of an ecosystem. Chipotle is using these insights to navigate a K-shaped economy..and how? Menu Changes: They aren't just raising prices; they’re introducing 𝐇𝐢𝐠𝐡-𝐏𝐫𝐨𝐭𝐞𝐢𝐧 add-ons specifically designed for that affluent 60% who are less price-sensitive. Resilience Testing: In other words, how much wiggle room you show in your spending with increased prices by testing 1-2% increase at a time or for trying new menu options in addition to your planned spending. That's predicting price elasticity. At the end of the day, we have to realize that every time we shop, we aren't just buying lunch. We are a walking, breathing data point for the companies. Is your lunch order more tech-forward than your actual job? What do you think? And did you think if you can limit companies do this to your data? opt-outs are there if you are interested...
2 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content