HN user
Has anyone tried acoustic imaging for water leaks inside walls? I live in a multi-floor 1900s Victorian. A leak can affect several units, and tracing the source can mean opening walls or floors in multiple places, and coordinating access has been getting harder with less WFH.
Could one of these tools help map water pipe routes and trace a leak, or are they only going to be useful for air and gas leaks?
My guess is the downtime is tied to the routines rollout https://code.claude.com/docs/en/routines
In my experience, knowing you have glaring UX problems, or that product does not have an easy intuitive user flow is rarely the bottleneck for developing new & useful user facing AI applications.
There’s usually a very real and very hard to describe data related impracticality that voids the usefulness of a design that appears well thought out and complete.
Additionally enterprise AI products are built on custom integrations, and complexity of maintenance overwhelms the engineering team and leaves very little time to build out new things.
The simplest changes that come from knowing insider customer experience have significant impacts. If the default range for a duration filter is 5-30min, and it turns out the most interesting data is really on 1.5hr+ rows. Or adding search across legacy platforms that bury uniform information under deeply nested modals, which people spend 20+ a week clicking through to collect a usable sample set based on existence of a few keywords. But building a system that returns good search results is the hard part.
I do like the “build on top” pieces in your gallery. If it’s fast and reliable enough to collab during a discovery meeting, or a customer success meeting, that would be genius. Because then you’d have a way to pull customers into the right mindset to articulate frustrations with their current software, iterate on getting those frustrations get translated into concrete designs together, and at the end you walk away with something that proves you both understand and can solve their problem to any audience.
There’s no point comparing apples to deep fried oreos for caloric density. The 919 Evo is a fully de-restricted prototype based off a legendary homologated race car, not remotely in the same category. The BYD U9 is a road-legal EV, comparing the two doesn’t mean much.
Funny you mention the Ford SuperVan because that’s much closer to the 919 Evo in the "no homologation no limits" category than anything you could register and drive off a lot. A fairer and much more impressive benchmark is the road-legal Ford Mustang GTD running a 6:52. That's still far quicker than the BYD, with roughly two thousand less horsepower.
Looks nice. For a true whiteboard experience, I think the 'Draw' tool should probably be the default rather than 'Select'. I was clicking around at first and couldn’t figure out why nothing was showing up.
The San Francisco marathon is this weekend and I think it’s going to break the city for this Robotaxi launch.
There's major road closures for key arteries like Market st, Embarcadero, fisherman’s wharf, and the Presidio. Traffic always crawls and downtown will become a maze. Even 'human' drivers struggle because you can't cross large boundaries of the city.
Waymo launched in the city about a month before last year’s race. I took one to the starting line, but it couldn’t reach the actual drop-off. It stalled about 0.3 miles away and I had to run the rest. The issue wasn’t the route, but the chaos. Dense foot traffic, impromptu street closure re-routes, and unpredictable crowd behavior were hard to autonomously solve.
Tesla's robotaxi launch will have to overcome the same challenging mix of realtime conditions: limited access to closure data, learning of impromptu re-routing logic, unpredictable human crowds.
Definitely it’s a bold move to launch this weekend. If it works, great PR.
Actually this sounds great. I got way more out of codecademy’s in-browser, interactive challenges than I did in my middle & high-school classes programming classes. The "learn by doing" process really built my confidence. If you could "demo" your docs directly in the browser it's much easier to learn by doing. I think that'd drive up adoption and you might even crowdsource bug discovery.
I used to have plenty of energy running 40-50miles per week, but when I ramped up to 80mpw I started nodding off in my chair by 1 PM. Then despite the higher mileage and more intense training, my race times slipped. My half went from 1:23 to 1:28, and I felt drained, irritable and angry unless I took a long break. After digging in, I learned that very high mileage can deplete iron levels. Once I focused on improving my iron absorption, I finally got my energy back and everything clicked. Even while holding 80+ mpw for the upcoming SF Marathon, I still knocked 5% off my Bay to Breakers time (48:51 this year) and cut my 5K PR from 19:17 to 18:37.
Curious how you landed on the idea to partner with local GCP/Azure reps. That’s a smart move, I didn’t realize they’d be open to helping. Did you pitch it as a way to help them close deals by offering custom solutions?
FWIW I built a streamlit app to extrapolate tribal knowledge in excel trackers into markdown wikis for vector database ingestion. Instead of uploading raw tables, it maps sheet headers to real headings to wrap each section in wiki-type format context pages. The UI lets you pick out QA sections from local files, but I’m stuck on how to persist selections and configs for repeat runs. Curious how others would tackle the issue of repeatable settings.
Code’s here: https://github.com/devin-liu/excel-to-markdown
This is fascinating. Could the same machine (PRAD) be adapted to zap mosquitoes in flight and help control disease vectors? Nothing’s worse than waking up to multiple mosquito bites and dreading falling asleep knowing there’ll be more.
The current title “How aging stem cells fuel visceral fat gain in middle age” suggests that existing stem cells simply ramp up fat production as they age.
Single-cell RNA sequencing of APCs then identified a new committed preadipocyte population that is age enriched (CP-A), both in mice and humans. CP-As displayed high proliferation and differentiation capacities, both in vitro and in vivo.
It should read more like "aging triggers emergence of a specific stem cell type (APCs) which drives visceral fat expansion"
disclaimer: not data or advice.
We once interviewed a former national [racket sport] champion from [country] for a React/JS SWE role. Our in-house [racket sport] expert, who happens to be the best player among us, was sure they’d ace the coding exercise. The next day, when we asked for his verdict, he gave us a few words: “He sucks. He got absolutely nowhere.”
Lesson learned: extreme talent in one domain not always a predictor of aptitude in another.
Any advice for leveraging a startup role to boost your career? I’ve seen friends turn their positions at early stage companies (Pre-Seed/Seed/Series A) into things like raising funding for their own ventures, hosting galas at SF museums, putting on international fashion shows with their alma mater, or even speaking gigs. And honestly, I have no idea how they pulled it off.
First off, that sounds incredibly frustrating. I’ve seen this archetype before—a business person with just enough technical knowledge to chime in on complex product decisions, but without the patience or interest to develop any real depth. Instead of building real understanding, they rely on mimicking what looks successful. Always chasing whatever has funding or buzz.
That kind of thinking lacks vision and makes it really hard to build anything meaningful. Over time, it wears you down. Even if the product “works,” it doesn’t feel good because you know you're forcing it. These folks tend to say things like “this should be easy,” then only show up to question why something isn’t done. They manufacture urgency instead of clarity, and when things inevitably fall short, the blame falls on the technical team.
Second, you’re not a first-time founder anymore. Life’s short. For me at least, you’ve earned the right to walk away and build something you actually care about again.
People have had similar sentiments in tennis about how racket and ball technology has changed the game over the years. Moving away from wooden rackets led to a massive increase in power and a larger sweet spot, which transformed the game from finesse to powerful serve-and-volley play. John McEnroe began with wooden racquets, while Pete Sampras and Andre Agassi adjusted to carbon fiber frames. Then poly strings took things even further, players generated extreme topspin to deliver aggressive swings with much more consistency, pushing the game back towards the high-powered baseline style.
For me, Roger Federer's style represents tennis at its most beautiful. His all-court game feels effortless and graceful, almost like a dance. But from a court-level view, it's more of a high-speed chess match built on calculated aggression, constantly pressuring opponents and waiting for the slightest opening to strike a point-winning shot. That level of sophistication and precision wouldn’t be possible without modern racket technology.
I still feel emotionally tied to classic matches from my childhood, especially Federer versus Nadal. But there's no objective reason, because tennis keeps getting better. People worried finesse was disappearing, but players like Alcaraz have brought back drop shots and clever cat-and-mouse tactics against deep-baseline defenders like Zverev and Medvedev. It’s a technique that was once considered too risky to rely on consistently.
In golf, tennis, baseball, basketball, running, & any other sport will keep evolving as technology & athleticism improves. Clinging to older styles feels more like holding onto the past than genuinely appreciating progress. If you can’t enjoy Curry hitting daggers in the Olympic finals or Kiplimo breaking 57 minutes in a half marathon, maybe the problem isn't with the sport itself. Maybe it’s the comfort of past memories holding you back from appreciating what’s happening now.
Yes, the customer is returning, but that’s completely normal in the bespoke tailoring process—it’s not new business. The process of getting a completely bespoke suit there's usually multiple fittings over several weeks. It's normal for the customer's body to change, and adjustments to made to create a better fit.
That’s why it becomes such an issue when customers come in requesting an alteration—it’s like being dropped into a team at the final stages of a project that leadership says is 90% done, but it’s been stuck for weeks trying to finalize that last 10% due to some "small last minute requirement changes"
I love my garmin, it just feels like the perfect fit for me. I have zero interest in the Apple Watch, mainly because I don’t want yet another device bombarding me with notifications. Plus, the Apple Watch just doesn’t seem like something I’d be comfortable getting sweaty and slathered in sunscreen every day. The rubber and somewhat industrial design of the Garmin feels like it's just made for running, not for juggling texts or emails mid-workout.
I first started tracking my runs with apple health, basically carrying my phone in my pocket to measure distance. Back then, I had no weekly mileage targets, or pace goals. Just a curiosity about how far I could run. Eventually, I switched to Strava. I felt a bit of friction around starting and stopping runs on the app, but I loved watching my paces gradually improve month by month.
Eventually I signed up for my first marathon, taking my iPhone in my pocket and first gen airpods that ran out of battery halfway through, but I finished in 3:48. I stuck with the iPhone for a while, but one day I zoomed into the strava map and realized the iPhone’s GPS was unreliable—it added zigzags to my routes, inflating my mileage and making me seem faster than I really was (massive ego bruise). So I went to research accurate GPS watches, and I remember seeing people test them by running straight lines to check for accuracy on a map. The forerunner was the most satisfying straight on the map, and so I bought that in May 2020.
So I’ve had a garmin since May 2020 and still love it. The simple start/stop mechanism has become a ritual for me. I also appreciate the heart rate screen, which shows my zone using colored ranges—it’s what I used to pace myself during races. For example, I’d aim to stay under 160 bpm during half marathons and marathons. With the Forerunner, I brought my time down to 3:11 for the marathon and 1:24 for the half marathon. That’s when I hit an inflection point: I couldn’t improve further without serious training plans.
I tried using Garmin Coach but made the mistake of choosing plans slightly below my fitness level. As a result, I didn’t run enough hard workouts and plateaued. After that, I lost motivation and took a break from running and lost fitness-- my old 130BPM pace became my new 160BPM pace. When I returned, I spent a year trying to regain it. I watched countless YouTube videos and read Reddit threads claiming, "every amateur runs too fast and too few miles." So I focused on high mileage without prioritizing aerobic envelope workouts. My fitness stagnated—my half marathon slowed to 1:27, and my 5K and 10K times didn’t improve. I also psyched myself by overshooting mileage targets, leaving me either sick or over-fatigued on race days.
Eventually, I gave myself permission to run hard again, and my fitness returned. I worked my way back to a 3:02 marathon last year. Now my favorite workflow involves using the VDOT app as my personal coach. I set a weekly mileage target, specify which days I can handle hard workouts, and it generates a detailed plan for me. For example: warm up for 2 miles, run 400m at a target pace of 5:40 with 1-minute rests, and cool down for 2 miles. The garmin integrates as what I call my "buzz coach" through each stage of the workout. Too fast? Buzz. Too slow? Buzz. Next lap? Buzz. The alerts really help with making real-time adjustments. Overall I find this setup eliminates the decision fatigue of training. I used to obsess over pacing, distance goals, and analyzing every bit of my data. Now it feels like I'm just getting outside, running a lot, and having fun with it—and ironically, I've just started improving again.
Curious—what useful things are you looking to work on right now? I'd love to learn more and help out.
In my opinion, this AI development stalemate is more layered. Big companies set such broad targets in a race to catch up with OpenAI that they lose focus on real use cases. So the loudest voices, those good at navigating internal politics, end up in a good spot to push their own ambition over actual customer needs or technical practicality. They set goals that sound just a bit more exciting than their peers, which pulls resources their way. But the focus shifts to chasing KPI's rather than drilling into real problems. Even when they know going smaller is smarter, knowing and doing are two different things.
It’s still a great time for small AI startups. My favorite kind is a team that quickly learn a business’s needs, and iterate toward the right interaction points to help. I think just staying focused on solving a lot of small related problems very fast, you can create something that feels like a real solution.
Have you looked into agentic chrome extensions like MultiOn? They use a similar class of AI model, but work on top of existing open browser tabs.
Could you explain more on how to do this? e.g if I am using the Claude API in my service, how would you suggest I go about setting up and controlling my own inference endpoint?
From my experience, ASR-to-NER pipelines don't perform adequately out of the box. Even though SOTA ASR systems claim 85% word accuracy, the distribution of errors is worth looking into. Errors around critical entities like credit card numbers or addresses are particularly prone, and even a small mistake renders the result useless.
These ASR errors cascade into the NER step, further degrading recall and precision. Combining ASR and NER into a joint model or integrated approach can reduce these issues in theory, it's just more complex to implement and less commonly used.
In the words of Gilfoyle-- I'll bite. Why has CUDA won?
I work great under pressure, but leaning on that long-term keeps me in a constant state of high stress, which messes with my baseline mood. I’ve found that clearing my head really helps, and I’ve been using the Sam Harris meditation app for that.
Most mornings, I’ll get a run in before work, and sometimes I do track workouts with my Garmin watch + VO2 app. The VO2 app has preset workouts e.g a warm-up mile, 4x400m at a target pace with 1-minute recoveries, then a cool-down mile. My Garmin tracks everything—the distances and the pace I actually ran—so I’m not having to think too much about it during the workout.
Afterwards, I get a summary with accuracy ratings on how close I was to my distance and pace targets. It’s awesome because I can track my performance over time and actually see that I’m hitting faster paces with less effort. That’s been great for building my confidence. I’ve been wanting something similar for desk work—a “workout plan” of timed laps to guide me through tasks and track my focus duration.
I’ve put together a rough version where I ask ChatGPT to make me a “workout plan” for desk tasks, and I made a lap-timer app on my Mac to parse it. Made a quick quick video showing an example for emails, though I'm not opening emails for the demo—just hitting the laps. https://youtu.be/tH7KFLC2640
If anyone knows of a better solution, I'm willing to try!
I’ve got a Duravit toilet here in the US, and I'm on the hunt for a compatible cold-water bidet seat under $300 on Amazon. I originally went for a Brondell, thinking it would match the shape and dimensions. But halfway through the install, I realized it wasn’t going to fit around the seat connectors. Duravit's own bidet seats are around $1,000-- I'm tempted, but they look like a lot of work to install. I’m starting to wonder if I should just upgrade the whole toilet to a model that comes with a bidet.
I’m not sure what’s behind this, could be endorphins, adrenaline, or something else entirely-- but I wonder if it’s related to that “runner’s high”. The first time I ran a 20-miler on a Sunday, about 3 hours of steady, low-key cardio, I went through an entire box of pasta afterwards and still felt like I needed more food right after. Now, after doing it 20+ times, I don’t even feel extra hunger kick in after the workout.
On days when I wake up and do something intense, like 4x1K repeats or 2x2-mile threshold sets, I feel electric afterwards, but at at the same time thinking about food makes me slightly nauseous. I’m not even close to hungry for hours, which throws off my eating for the rest of the day.
But then there’s carryover calorie deficit. Let's say the next day I wake up for an easy day with a 5-8 miler, I’m sold on breakfast right afterwards, and I can usually eat any amount that's in front of me.
Respectfully, what experience or expertise backs your statement? As someone who has personally faced obesity and struggled with weight management over the years, I have a somewhat different perspective. Growing up, I was an obese child and even made the varsity tennis team as an underclassman at an obese BMI. I tried dieting, cardio, and nothing really helped me lose weight. I eventually found some success with a keto diet before my senior tennis season. However, I found I had to start eating carbs to stay competitive, which led to weight fluctuations but also better performance.
Even now, as an adult, I find weight management complex—I've been close to obesity while running up to 80 miles a week in marathon training, hitting a 3:02 marathon (6:58/mile pace). After finishing the marathon and cutting back to 40-50weekly mileage, my weight just naturally decreased. My appetite was much less when I wasn't running such high mileage. For me, it's a journey that seems to involve many factors beyond just low physical effort or overconsumption.
Screenshots aren't as accurate or context-rich as HTML, but they let you bypass the hassle of building logic for permissions and authentication across different apps to pull in text content for the LLM.
but the ultimate goal is to crowdsource a high quality set of browser sessions to train an open source local model.
Could you say more on this? I see that it's an open-source implementation of PLAN with Selenium and Claude's Cursor, but where will the "successes" of browser sessions be stored? Also, will it include an anonymization feature to remove PII from authenticated use cases?
I recently had some luck turning an excel tracker that lists multiple locations and their services into markdown for RAG. It worked great as a natural language lookup, way better than digging through a big Excel sheet.
I uploaded them through Supabase Embeddings Generator if you're curious. https://github.com/supabase/embeddings-generator
But things got a bit messy when I handed it off to someone else. They started using synonyms for locations, like abbreviated addresses to refer to certain columns, which didn't return the right documents.
Followed a friend's suggestion to try NotebookLM, so I uploaded the same docs there, and it was awesome. Some cloud-hosted vector DB tools only handle PDFs, but NotebookLM accepted my Markdown and chunked the docs better than the Supabase library I was using. It just "worked".
I would swap over to NotebookLM because their document chunking and RAG performance is working for my use case, but they just don’t offer an API yet.
I also gave Gemini a shot using this guide, but didn’t get the results I was hoping for. https://codelabs.developers.google.com/multimodal-rag-gemini...
Am I overhyping NotebookLM? I’d love to know to get on-par document chunking, because that seems to deliver fantastic RAG right out of the box. I’m planning to try some other suggestions I’ve seen here, but any insights into how NotebookLM does its magic would be super helpful.