|
Hi, I am itching to work on some side projects and thought I might blog about them in BFC as they're generally all business ideas. I posted 4 ideas below and will probably start working on one maybe 5-10 hrs/wk in 2-3 weeks. Vote on your favorite idea so I can ignore it and become the goon in the well! I am leaning hardest to #1, currently. Idea #1: List coins on eBay with two pictures. quote:What problem(s) does this solve? Making eBay listings takes way too much energy to make it worth listing coins selling for $10 or less. To list a coin you need the type, grade, date, mint mark and some photos at a minimum. Its hard to get more than 10 listings up per hour. Shipping takes time as well, but coins have fairly standard weights and sizes, potentially enabling a “ready to ship” packaging method. Additionally, it takes expertise to grade coins that most people do not have. Idea #2: DollarProtest.com quote:What problem(s) does this solve? Boycotting a company or protesting can only work if your voice is heard. Social media has become a platform where anyone can complain, but rarely are these taken seriously. A snail mail letter each month is a physical reminder that you’re unhappy. Idea #3: Automatic shower cleaner quote:What problem(s) does this solve? Cleaning my shower sucks. Idea #4: Simple Factory quote:What problem(s) does this solve? Building an automated factory is too hard and prototyping is too expensive. This difficultly prevents otherwise enterprising people from solving challenges which require bespoke machinery. Request: If any of these ideas get any traction you'll be able to know who I am. Please don't dox me here or elsewhere. What do you think? Which would you use? CarForumPoster fucked around with this message at 16:43 on Jun 29, 2021 |
#
?
Jun 29, 2021 15:06
|
|
|
|
| # ? Apr 26, 2024 08:26 |
|
|
#1 seems like the best option. You can just do it. Don't need people/capital so weird and niche you're going to be "the guy" Plenty of potential. Expanding into similar categories is prob just training a new algo? If you do really well maybe ebay buys it from you. Making the marketing/growth side easier because you just need to show promise. Plus real $ at the end. Shower thing sounds cool but man ... hard to really get into physical goods as a one man operation. You at least need a Chinese uncle or something to produce your first 1000 copies or something. Xguard86 fucked around with this message at 16:01 on Jun 29, 2021 |
|
#
?
Jun 29, 2021 15:59
|
|
|
CarForumPoster posted:Hi, I am itching to work on some side projects and thought I might blog about them in BFC as they're generally all business ideas. I posted 4 ideas below and will probably start working on one maybe 5-10 hrs/wk in 2-3 weeks. For Idea #3: Automatic shower cleaner I was going to say there is actually an option, but reading up on it the scrubbing bubbles auto shower thing was discontinued. And apparently it's a rather hard problem to solve, who knew?
|
|
#
?
Jun 29, 2021 16:25
|
|
|
Xguard86 posted:#1 seems like the best option. Yea that's my thinking. And if it made $50K/yr in profit without much work to maintain it I'd call it a huge success. I've always wondered why eBay doesnt do niche listing helping apps like this. While the solution I have doesn't generalize, you'd think that they'd see enough value in listing an grading Hummels, pokemon cards, whatever to make a simplified workflow for these. That said, eBay strikes me as the least innovative tech company in the valley. Deeply entrenched brand monopoly from the pre dot com era. Good for them, but bad for innovation. That's my feeling on the shower thing too. Cash flow with 2+ month procurement times and a high risk of defects is a real bastard. And the thing that stops humidity in this design is workmanship on the potting. I've imported parts machined to my spec from China back in the late 00s, though never electronics. I definitely have some Chinese speaking friends that would travel to China with me though if I paid their ticket. downout posted:For Idea #3: Automatic shower cleaner I was going to say there is actually an option, but reading up on it the scrubbing bubbles auto shower thing was discontinued. And apparently it's a rather hard problem to solve, who knew? Yea Idea #3 exists because I tried to buy that shower cleaner. An ex had one and it seemed decent. I've seen then insides, the design is dog poo poo. That said, I'd prob need to sell mine for ~$75-100 and I'm not sure what the markets apatite for it would be in that range. CarForumPoster fucked around with this message at 16:33 on Jun 29, 2021 |
|
#
?
Jun 29, 2021 16:31
|
|
|
#4 sounds fascinating tbqh, and most likely to just get bought up by some large company if you have a few patents. I don't know enough about the industry to say more but it sounds like a fantastic possibility. #2 is going to be tough. You'll either have to eat a painful fee for every months' charge, or schedule a 6-12 month spend. From an efficacy perspective, you'd probably be pissing into the wind for all the effect of it. Far more effective is joining a particular interest group like The Sierra Club or something; I'd actually love to know from someone in even a moderately large organization how much attention they pay to mail sent; I assume anything with AUTO on the postmark will just get thrown into a junk bin. #3, the shower cleaner sounds like it would be pretty rough to design at a price point that people would actually pay. I'd love to have something like that, but paying $100 for the razor made by the guy whose blades may not even be around in 3 months is going to be a big stopping point. #1 is interesting, but I also don't know how much volume coin collecting on eBay would do. Definitely sounds like a possible niche, especially if you can be the primary developer. Concerns I can think of: 1) incentivizing new users without eating a ton of cash on paying for ML service use from end users that don't end up completing the flow 2) preventing abuse, you will need to provide a way to manually correct incorrect results but need to deal with scammers intentionally using your platform to provide fraudulent or misleading listings (what do you mean it's a X cent piece? MyCoinBay told me it was a Y, that's why I listed it as that!)
|
|
#
?
Jun 29, 2021 23:10
|
|
|
Volmarias posted:#4 sounds fascinating tbqh, and most likely to just get bought up by some large company if you have a few patents. I don't know enough about the industry to say more but it sounds like a fantastic possibility. Thanks! Re #3: I wouldn't follow SC Johnsons pricing scheme for the shower cleaner, I dont think. I'd just make the container refillable, you can put whatever in it and I'd have a formula under whatever brand name its made as also. Re #1: Last year I estimated there's roughly $12M/yr of market volume in American coin auctions under $50 on eBay that actually sell.
|
|
#
?
Jun 30, 2021 01:24
|
|
|
I spent a few hours today on the coin project. I was testing tools for bounding box labeling and semantic segmentation to try to get a way to find the dates and mint marks. I found this pretty sweet, free tool: https://www.makesense.ai/ ...then I realized that this is probably not a problem I need to solve right now. I can have some Amazon Mechanical Turk people review and draw boxes on my actual images for such a low price per auction ($0.20/ea) that it would likely make more sense to just have them do the labeling for me and I review it. This lets me bypass the whole "make an AI find the date and mint mark" obstacle. The MTurk API sucks, but I've used it before and can make a review app in a day or two. I'd estimate less than 40 hrs to get the whole mturk workflow up and running including instructions for them on where to look for each type of coin. The biggest obstacle to using an MVP on my personal collection is now some automated pricing tool so I know which coins to list and which to put together in lots. I'd ideally still like to have a photo for each coin in the lot because why not, right? I think I'll work on this next.
|
|
#
?
Jun 30, 2021 01:35
|
|
|
#4 isn’t doable, if it was it would already have been done by a mega Corp. looking to save a few bucks. I mean a car is mostly built with wrenches and screwdrivers but it isn’t with the type you buy at Home Depot for a couple of bucks.
|
|
#
?
Jun 30, 2021 01:57
|
|
|
CarForumPoster posted:The biggest obstacle to using an MVP on my personal collection is now some automated pricing tool so I know which coins to list and which to put together in lots. I'd ideally still like to have a photo for each coin in the lot because why not, right? I think I'll work on this next. Can you get started with some kind of static index from the latest "coin collector monthly" or something?
|
|
#
?
Jun 30, 2021 02:05
|
|
|
Xguard86 posted:Can you get started with some kind of static index from the latest "coin collector monthly" or something? Yea but there's always the issue of usage rights. That said, I have access to the eBay developer API, I've chatted a couple times with the CEO of CPG Greysheets, and this exists: https://www.pcgs.com/prices/us so I can def cobble together a legal and "useful to start" way to use pricing to make decisions on "goes in a lot" versus "list by itself". Long term I prob want to make some sort of regression model for pricing that I keep updated with the latest data. Implementing that in code prob won't be too hard but we'll see. CarForumPoster fucked around with this message at 14:36 on Jun 30, 2021 |
|
#
?
Jun 30, 2021 04:14
|
|
|
sbaldrick posted:#4 isn’t doable, if it was it would already have been done by a mega Corp. looking to save a few bucks. Mega corps do this. An auto company's engine plant I toured is filled with OKK horizontal mills, pallet changers, in process measuring devices, etc. all integrated with each other and recording data via several sensors. They use the same machines and pieces over and over. Prior to installing these, the floor and process is laid out by a team of process and automation engineers. They'll do simulations of the throughput, system design, placement, communication tests, etc. as the lead time for procurement are long and the costs of delays are high. They don't need to worry as much about adding layers of abstraction to the actual device integration though because of the team of engineers. Another example more in line with what I'm picturing: Any big defense co that builds planes containing mission computers will have servers that simulate the mission computers, effectively simulating the hardware without buying pieces, then a systems integration lab that simulates the integration within the plane, but with much cheaper networking and better reconfigurability. Some will even go beyond that and have sections of the aircraft in there so you can really test how it'll be set up. Those systems have many standard interfaces, particularly for networking. EDIT: The use of open architectures in the defense industry is written into U.S. law. The problem with manufacturing IMO is it is filled with non-networked, expensive hardware and proprietary interfaces. If those were replaced with open architectures, you'll end up with something more like a plane manufacturer's integration lab. CarForumPoster fucked around with this message at 05:37 on Jun 30, 2021 |
|
#
?
Jun 30, 2021 04:54
|
|
|
TRIPPPPLE POST Looks like at least one dude has taken a crack at the AI grading idea, though it appears to be not released and even the identification feature appears to not work. Using it is fairly painful. Takes him 10 minutes to grade 2 coins lol. https://mintstate.com/ Here's him trying to use it: https://www.youtube.com/watch?v=fhWq_auONm0 CarForumPoster fucked around with this message at 14:34 on Jun 30, 2021 |
|
#
?
Jun 30, 2021 14:24
|
|
|
CarForumPoster posted:TRIPPPPLE POST Well that's good news. You're on to something if someone else tried it and you have an opportunity if their version kind of sucks!
|
|
#
?
Jul 1, 2021 01:40
|
|
|
Xguard86 posted:Well that's good news. You're on to something if someone else tried it and you have an opportunity if their version kind of sucks! Hear! hear! Today I thought through the process on how I can do the pricing bit so I can decide what goes in a lot and what goes to auction by itself. Which made me realize that without a date, which was the point of the mturk people in the first place, I can't really decide what something is worth. Key dates can be worth a lot more. Anyway I'm thinking I am back to making a model to draw boxes on dates and mint marks. No bother, won't be too hard. CarForumPoster fucked around with this message at 02:09 on Jul 1, 2021 |
|
#
?
Jul 1, 2021 01:49
|
|
|
I was reading the hobby CNC thread thinking on SimpleFactory (idea #4) and where the most friction is to making a factory. I think the biggest mental hang ups for me have been -getting the motion control portions working. -getting the machinery precisely calibrated. Like a specific process to test and adjust for squareness, flatness, etc. I think people dont feel as daunted by the programming aspect, if they could know that plugging certain things together and running a process would "just work" and not have 9 million little issues to debug or rebuy. There's just SO MANY options and its hard to know what size of everything you need. Also, it seems like literally every motion system project runs into EMI and/or connector issues. Makes sense because there are long runs of often heavy wire and dealing with noisy motors and cheap electronics. However most of the mitigations for these are easy if you're using standardized stuff and plan ahead for EMI best practices. Anyway, maybe where I need to start with SimpleFactory is: -a pretty, visual online configurator to let someone daydream about what they want to build. Like what the auto industry uses. Have templates for projects that are common like lathe or milling machine conversions. -a machine controller that can interface with the most widely used motor drivers (for steppers and servos) trivially via pre wired connectors and which is trivially easy to communicate with. Make it so you can upload the online configurator's output to get the basic system up and running. -an online store that sells the pieces for the configurator or has links directly to the products needed. -videos and instructions on putting it together (incl EMI and wiring instructions) -videos of us building equipment in just a few days which we then sell CarForumPoster fucked around with this message at 20:15 on Jul 1, 2021 |
|
#
?
Jul 1, 2021 18:56
|
|
|
Some thoughts - #1: You could, as a follow-up, provide some way to ship them. Ie sell custom shipping boxes etc, or a specialized Pirate Ship . But perhaps this is out of scope. And maybe there's a more targeted way to handle this than deep learning, but I don't know what that is, and it sounds like you've already solved this problem. Make sure there's a confidence indicator that will revert to "manual" control if below a thresh. We've talked about #3 in depth! Go for it. Most people will say they have no use for it, but you don't need most people to buy it. #4 As you posted, this is a very hard problem, and the implementation details, audience, and cost will vary significantly on what sorts of devices these systems are capable of producing. You're likely to run into many people telling you this is impossible. Fits under the category of "cool sci-fi things we don't have, but could". Easily the most ambitious of what you've posted. Go wild, and don't let curmudgeons used to doing things in a specific way discourage you. You'll do well starting with a very confined scope, and expanding from there. This task may be beyond your organizational or funding abilities, but I think we'll see a lot of progress on it over the next century. You should make #1 and #3 happen. Then move onto #4 later if you're feeling ambitious, and are prepared for failure. Dominoes fucked around with this message at 01:15 on Jul 2, 2021 |
|
#
?
Jul 2, 2021 01:12
|
|
|
Dominoes posted:Some thoughts - The beauty of an ebay coin app is cash flow with low/no time commitment once you build the thing. It's just pure glorious profit margin. I, being a dumb engineer who wants to build physical stuff, want to take the app and make it a service where I photo, list and sell the coins with a little automated machine that drops the coins down, packages them for shipping with the only human bit being loading them into the queue. I've actually designed a whole machine for this purpose and everything including a rough prototype. There's a bunch of logistics with that though where as a coin app that grades and lists is just like...sign in through the eBay auth mechanism, you first 5 listings are free. Have at it. I wouldn't worry too much about the funding and resources part. Raising seed money is the easiest part of starting a high growth startup. For SimpleFactory if I bootstrap a launched product with some user feedback I bet I could go through YC again, raise $1-2M at a 10 cap with no problem. Especially if I can minimize the amount of hardware I actually build ourselves. Many of the benefactors of the SV exits are engineers too. Like us, they have a boner for future building products that function as platforms and are "ultra sticky". That said, I still am working on our current YC startup so I'm more looking to scratch smaller problems and maybe I'll build something that needs SimpleFactory...ya know...like the coin machine or the equipment to make the shower cleaner haha.
|
|
#
?
Jul 2, 2021 02:49
|
|
|
CarForumPoster posted:Idea #4: Simple Factory quote:The problem with manufacturing IMO is it is filled with non-networked, expensive hardware and proprietary interfaces. Hardware is expensive for a reason; it needs to be extremely robust because downtime costs money. An arduino isn't going to stand up to a manufacturing environment and when it goes down it will only take an hour before the downtime has cost more than the price difference. If you're interfacing to an existing machine, there is going to be at least some way to connect (RS232, EIP, Profinet, Direct IO, etc) and manufacturers are indirectly incentivized to assist in integration. quote:A method to simulate the system so you can build it in code + CAD before buying components. quote:I think people dont feel as daunted by the programming aspect, if they could know that plugging certain things together and running a process would "just work" and not have 9 million little issues to debug or rebuy.
|
|
#
?
Jul 2, 2021 04:56
|
|
|
TheLastManStanding posted:This is extremely vague and I'm not sure what your actual goal or product is. It sounds like you want to make a general solution to something, but a general solution will never be able to compete against a highly specialized one in a high production environment. I mention safe twice in the OP but "vaguely defined system with unclear goals that doesn't have basic functional requirements" is a correct summation of the state of this idea. Back in 2009~2014 I used MasterCAM a lot for CNC programming and was trained on (I think) Robotmaster to program a Kuka our shop had but never really got to use it for anything beyond one complex sanding job. I am, or was, a machinist with 10 fingers so I'm all on board for the importance of safety. Visual Components looks awesome. I think where this idea needs to go is into practice. Maybe I'll build the automatic coin listing machine and then retrofit a lathe to have a through spindle bar feeder. Buy some old CNC mill with a dead Fanuc control and retrofit it. If I do those three and say "boy this was easier than expected" maybe the market has in fact solved this problem and I just haven't surveyed it properly. EDIT: BTW your post spurred me to look for what has come to market for CNC motion controllers and it looks like there HAVE been several developments I haven't seen before so thank you. For example there are low cost single axis controllers like this one. I bet there are some that can take a drip feed of G Code or some other "buffered" set of commands. (i.e. they don't rely on a real time communication from the driving computer, thus requiring an FPGA-based motion board or something similar.) And I bet there are some that can accept a bunch of input for various sensors or control of relays for coolant pumps and what not. I think I might just be behind the times here. EDIT2: Re: Proprietary interfaces. I'll give a real example of the perceived friction. In 2014 I had a Fanuc Robodrill w/24K spindle with a Renishaw probing setup. This is a sweet machine, easily $100K. It was maybe 2 years old. -Supposedly it could be fed programs over the shop ethernet direct from MasterCAM though, TBH, it was often easier to load it using the PCMCIA card. That's right! In TYOOL 2012, it was using a PCMCIA card for memory. Impossible to upgrade as it was built in to the control pendant. -Let's say I wanted to monitor spindle load on that machine for two of my tools to know when something went wrong. Well there's no API into any of the machine controls. -How about getting those probe measurements into excel for the customer? Nope, it'll display them on the machine because I can save them to macro variables, but Fanuc doesn't let me read the macro variables over ethernet. -Alright, maybe I want to add a tombstone. Well I better be ready to fork over big $$$ for Fanuc's setup, of which they offer one size that eats a ton of work envelope. It'd be better if the control was open to other developers. So yea sure they might be "networked" but they tremendously restricted what can be measured or added on to the machine, greatly reducing its potential capabilities. CarForumPoster fucked around with this message at 12:42 on Jul 2, 2021 |
|
#
?
Jul 2, 2021 11:40
|
|
|
The only, only factor that matters when speccing components for industrial machinery is that the supplier is large and stable enough that spare parts can reasonably be expected to remain available 20 years down the road. Also I've got a bunch of Robodrills and Index MS22's running Marposs Artis so not sure what the spindle monitoring dig is about shame on an IGA fucked around with this message at 14:48 on Jul 2, 2021 |
|
#
?
Jul 2, 2021 14:45
|
|
|
also plugging $3mm machine tools that still ship running WinXP in tyool 2021into the network should be, shall we say, discouraged
|
|
#
?
Jul 2, 2021 14:52
|
|
|
CarForumPoster posted:I think where this idea needs to go is into practice. Maybe I'll build the automatic coin listing machine and then retrofit a lathe to have a through spindle bar feeder. Buy some old CNC mill with a dead Fanuc control and retrofit it. If I do those three and say "boy this was easier than expected" maybe the market has in fact solved this problem and I just haven't surveyed it properly. quote:Re: Proprietary interfaces. I'll give a real example of the perceived friction. quote:Alright, maybe I want to add a tombstone. Well I better be ready to fork over big $$$ for Fanuc's setup, of which they offer one size that eats a ton of work envelope. It'd be better if the control was open to other developers.
|
|
#
?
Jul 2, 2021 16:47
|
|
|
This seems to have shifted into a debate about how open CNC machine controls can be/are and what the needs of machine shops are, so I'd like to step back. My goal with SimpleFactory isn't to make another unsuccessful CNC retrofit company, nor to support machine shops. If a business needs a CNC mill Haas has great financing options. I want to make it possible for motivated, creative people to build the first version of an automated factory to do something they currently don't believe they could achieve. Generally, that means turning stuff on and off, reading sensors and moving heavy (relative to hobbyist stepper motors) things around. IMO, turning stuff on and off and reading sensors is already really accessible. There's a Youtube video about every microcontroller, sensor, etc. and tons of well defined, open interfaces. For example: if they want to recycle paper in Utah, build their own pulp mill that makes egg crates for ostrich eggs. Or they want to build an automatic coin grading machine. My hypothesis is that moving heavy stuff around is where the problem feels unapproachable to most people. Hence the interest in motion controls as a place to start. CarForumPoster fucked around with this message at 19:59 on Jul 2, 2021 |
|
#
?
Jul 2, 2021 19:54
|
|
|
CarForumPoster posted:I want to make it possible for motivated, creative people to build the first version of an automated factory to do something they currently don't believe they could achieve. My main concern is how general-purpose you can make the process. Perhaps you should write down a list of specific requirements re scope. Dominoes fucked around with this message at 20:06 on Jul 2, 2021 |
|
#
?
Jul 2, 2021 20:04
|
|
|
Boring blog: No progress for two weeks as work picked up and spent time with friends. I did repaint my home office though. Planning to spend some time on coin app on Friday, workload permitting. Goal will be to build the pricing part of the app workflow. EDIT: Oh and I learned an easy way to start a dead car if you can't get a jump. I completely discharged the battery on my 2015 STi this morning by leaving a dome light on for 90 hours. Benchtop power supply current limited to 5A, voltage limited to 14.5V charged my car battery enough to start with no issue in about 20 minutes. I waited until the voltage dropped to 13.2V on the PSU readout and it started right up. CarForumPoster fucked around with this message at 19:49 on Jul 12, 2021 |
|
#
?
Jul 12, 2021 15:29
|
|
|
Got some coin pricing data together. On the todo to get the pricing part done: 1) Associate the ID numbers I have with the ID #s from the pricing data 2) Integrate the price lookup into the process of taking the photos. 3) Deploy an app that someone can use, even if its its type, grade, etc predictions are wrong. I'll just fill a dropdown with what it thinks the answer is and let someone correct it.
|
|
#
?
Jul 13, 2021 18:22
|
|
|
Here's some screens from the app, lmk your thoughts. Home screen:  Click + to add a new coin, which brings you here:  Once you add a coin it will be "Processing" for approximately 10 seconds. After it finishes, the name, date grade and value will be filled in and it can be accessed from the home screen:  
|
|
#
?
Jul 14, 2021 16:49
|
|
|
Looks pretty good. Small suggestions: change "processing" to "identifying"? A little more descriptive of what's happening. I see those letters S, O etc and figured out on the second screen that they correspond to the mint. Is that something a numismatist would Intuit? Idk the domain but might confuse people if it's not a super common thing. What does the camera/scan part look like? Idk how much you can control that VS the OS but a lot of keeping people engaged will be about the scan. Xguard86 fucked around with this message at 18:12 on Jul 14, 2021 |
|
#
?
Jul 14, 2021 18:10
|
|
|
Xguard86 posted:Looks pretty good. This is great feedback! Agree on processing, updated to Identifying. A numismatist would but I want this to be for everyone. what if it was "Mint: S - San Francisco" or "Mint: O - New Orleans"? The camera would be the OSes native camera, I wouldn't have much control in version 1. I am considering, for quality and accuracy, having the user print out a sheet of paper that has markings and orientation on it. Several coins are basically identical on one side except maybe for diameter which can be hard to tell with no frame of reference. What do you guys think of a "card" style design rather than the above list? Card Style - vs - List Style 
|
|
#
?
Jul 14, 2021 18:45
|
|
|
Even just appending mint: mint: S and mint: O would probably solve it because you're putting me the right context. It's such a balance, too wordy and it's actually less clear. So I'd say as little text as you can get away with. Lol let me be a true PM and say, I like both card and list. Probably should allow people to toggle. But you might defer adding that if it's a lot more complexity to build. Re: camera and guide. Yes! That is what I was thinking. people take good pics = better matches = more satisfaction = higher useage and stickiness. A cheap way to improve that loop is guiding the human element to take better pictures. If it could go inside the app all the better but if your heads in that space I bet you have good ideas over time. A printable scale is a good easy start. gently caress, this is 10x more fun than my day job.
|
|
#
?
Jul 14, 2021 19:18
|
|
|
Card style is stylistically better, but list style is going to be more usable. Do both, default to card style.
|
|
#
?
Jul 15, 2021 01:04
|
|
|
Volmarias posted:Card style is stylistically better, but list style is going to be more usable. Do both, default to card style. I'm using a no code PWA for version 0.1 so I am fairly limited. (For front end, backend will prob be an AWS Lambda function). This cuts my development time of an MVP dramatically though. I can get feedback and have a new version out in a couple hours. Tentative roadmap is this: I'm hoping within 2 weeks to show a few people version 0.1 of the app. It will take a picture and, within 10 seconds, spit out a mostly accurate type, within 1 category grade and a probably wrong date and mint along with an estimated price. If feedback is good but they're annoyed at the inaccuracy of it, V0.2 will fix that. Probably through the use of humans in the loop. I can likely bang that bit out in another 10 hours of work (2 weeks) and V0.2 will enable someone other than me to use it (add authorized users). If that gets built and people like it, but want to list on eBay I'll figure out the eBay auth workflow, start integrating other feedback and that can be V1
|
|
#
?
Jul 15, 2021 16:25
|
|
|
https://www.youtube.com/watch?v=oD17uLWd2qA
CarForumPoster fucked around with this message at 04:21 on Jul 16, 2021 |
|
#
?
Jul 16, 2021 03:40
|
|
|
Making progress on coin app, need to start thinking of a name. I got the deep learning models for the V0.1 REST API running locally using a project that is already deployed. Running locally it gives coin type, date, mint mark and grade in about 6 seconds, a good sign. However, Pytorch+deps are nearly 1GB by themselves. AWS Lambda limits you to 250MB. I'll probably need to change the way I am deploying which is a lot of effort for such a silly issue like file size limit.
|
|
#
?
Jul 16, 2021 19:13
|
|
|
Unclear if you're humble bragging about your billions or
|
|
#
?
Jul 16, 2021 19:44
|
|
|
CarForumPoster posted:However, Pytorch+deps are nearly 1GB by themselves. AWS Lambda limits you to 250MB. I'll probably need to change the way I am deploying which is a lot of effort for such a silly issue like file size limit. Lol, Don't suppose you could break anything up? Make several smaller calls? Maybe equal effort to changing how you deploy, or that's what you mean. Idk There's prob a good name in the idea that coins have faces... Literally portraits and head/tails. I'll post if something comes to me.
|
|
#
?
Jul 16, 2021 22:04
|
|
|
Xguard86 posted:Lol, Don't suppose you could break anything up? Make several smaller calls? Maybe equal effort to changing how you deploy, or that's what you mean. Idk Flip, since you flip coins but this also means reselling I will graciously accept 1% equity for this brilliant idea
|
|
#
?
Jul 16, 2021 22:08
|
|
|
Volmarias posted:Flip, since you flip coins but this also means reselling okay but im gonna dilute it first chance i get flip is also what you put coins in btw Volmarias posted:Unclear if you're humble bragging about your billions or my wife makes money, i make bad ideas Xguard86 posted:Lol, Don't suppose you could break anything up? Make several smaller calls? Maybe equal effort to changing how you deploy, or that's what you mean. Idk I've deployed very large things to AWS Lambda, e.g. web scrapers and their associated binaries, its just that the project I decided to piggy back on so that I didnt have to make a bunch of boiler plate bullshit for deployment is deployed with Zappa and I am developing on Win10 because I am bad at this. Well Zappa's large file handling methods dont work on Win 10 (and also don't work well in general). The thing I should do is use AWS SAM which is what I use for my other >>1GB projects but that means standing up a whole new stack and I only had like 5 hours to work on this. I could try to kludge together getting certain packages to be brought in from elsewhere, could install windows subsystem linux and see if zappa is able to make it just work, and/or some other hack but honestly I prob just need to do something that can handle large files trivially. As a reminder this problem exists because Amazon decided that 250MB was plenty.
|
|
#
?
Jul 16, 2021 22:42
|
|
|
Volmarias posted:Flip, since you flip coins but this also means reselling I actually kinda liked this so I looked and I can register fli.pa for $300 but not coinflipa.com and you really want the .com. Flipa is short and doesnt mean much and coin flips are a thing and it also means trading and it has that distinct urban patois that is popular with...numismatists EDIT: THEcoinFLIPA.com is available tho.....oh boy big money sitting on that domain OOH OOH coinflipper.life? huh? seems pretty great. Or maybe coinflipblog.xyz? theflip.money? CarForumPoster fucked around with this message at 22:57 on Jul 16, 2021 |
|
#
?
Jul 16, 2021 22:51
|
|
|
|
| # ? Apr 26, 2024 08:26 |
|
|
Anything with flipacoin? Flipacoin.now
|
|
#
?
Jul 16, 2021 23:06
|
|
