In the past years, I had a few different and weird business ideas / side hustles that I started, but in the end, many of them failed. Recently I decided that I’ll write a blog post about each one, mostly for myself to understand what went wrong and what I could do better in the future. They failed but were just fun to do and I got a lot of experience and good memories. From each one I learned something, but also I can see some common mistakes repeated each time.

Back In 2015, I started a small business – a website (twoja-foremka.pl AKA your-cookie-cutter), where you could upload your custom shape (PNG file) and I printed a cookie cutter from it in 3D. It was my third business about 3D printing that I started – I’ll write about them in the next posts.

How it was working?

All were semi-automated. I wrote a script that from the image file (png) was generating a 3D model suitable for a 3D printer. Then I was sending the model over email to someone that I found on the internet who was printing it and sending it directly to the customer.

I didn’t have many customers, just a few, but the margin was nice. I charged $30 for one, and printing cost me about $4, shipping $1. So I had $25 in my pocket – just for putting the PNG to the script and sending an email with an STL file to someone who prints it – about 5-10 minutes of work. For more technical details scroll down below to the last section.

What was done wrong?

I did have a 3D printer (RepRap) so I tested it myself and I was making cookies for the first time in my life. That was the first problem, I didn’t know anything about making cookies. But it turned out easy (thanks mom for helpšŸ˜‚) and here are some photos of cookies I’ve made while testing:

The biggest issue was that I didn’t have a clue who my customer was. I 100% focused on the technical side – which was fun for me. It’s not a business mindset. I was thinking that those cookie cutters are so cool because you can make a cookie from anything, like company logos or 3D puzzles, chess, lions, houses, etc. Imagine inviting your client to your office, and the cookies on the table are logos of their company – that’s brilliant. So I thought that maybe It’ll go viral somehow.

I didn’t write to companies or talk to the right people, I just set up Google / Facebook ads and I thought it will do the job for me, but it didn’t work this way. I also had a Facebook page (still online https://www.facebook.com/twojaforemka) and an online shop (PrestaShop), with ready-to-buy cutters, but with no traffic. So as I finished the interesting – technical side, I got bored and just left to the next idea – another big issue when you start a business.

What could be done better?

I should have found a non-technical partner that knows what to do to find customers and research the market. I didn’t have any motivation to do this. When someone says “marketing” or “sales” I just get instantly bored, I just can’t do this it is so boring for me. So I think the best thing I should have done is find a partner. But where or how can someone like this be found, someone that you can trust? If you know, post a comment, or write to me directly on my LinkedIn profile. Or maybešŸ¤”.. maybe this is a good business idea! A portal that combines technical people like me with people of marketing/sales?

What I’ve learned?

Aside from the issues above and technical stuff, the thing that had the biggest impact on me was the margin. I never thought that I could set it so high. It was shocking for me back then, that I can sell for $30 with costs around $5, with almost any work within the order.

Technical details – how I was converting PNG to a 3D ready-to-print STL file?

I already had some experience with Blender and using it programmatically with Python scripts. The mechanism was really easy. First, the PHP script was preparing the image, generating a stroke on it. Then the commands were run:

  • mogrify -flop $filename
  • convert $filename -transparent white $filename
  • convert $filename -transparent black $filename
  • convert $filename -fill black -opaque red $filename
  • convert -flatten $filename $filename.ppm
  • potrace -s $filename.ppm

Then the image was opened in Blender, and the script there was doing the magic. It was loading the SVG and extending the mesh upwards. Then it was exporting it to the STL file that can be sent directly to the printer. There was also a mechanism for creating 3D renders. There is the Python code for generating the model. I doubt it works now, as Blender often changes its API, and it is from 2015. I haven’t tested it now either.

import bpy, sys, copy, mathutils,time,math,array,bmesh,math


args = sys.argv
argv = args[args.index("--") + 1:]

def getArg(index):
    if (len(argv) <= index):
        return ""
    else:
        return argv[index]

def enterEditMode():
    bpy.ops.object.mode_set(mode='EDIT')

def enterObjectMode():
    bpy.ops.object.mode_set(mode='OBJECT')

def selectAllObjects():
    bpy.ops.object.select_all(action='SELECT')
    
def deselectAllObjects():
    bpy.ops.object.select_all(action='DESELECT')
    
def selectMeshAll():
    bpy.ops.mesh.select_all(action='SELECT')
    
def deselectMeshAll():
    bpy.ops.mesh.select_all(action='DESELECT')
    
def getAllObjects():
    return bpy.data.objects

def getMeshObjects():
    result = []
    for item in getAllObjects():  
        if item.type == 'MESH': 
            result.append(item)
    return result
        
def getSelectedObject():
    return bpy.context.scene.objects.active

def setSelectedObject(obj):
    bpy.context.scene.objects.active = obj
    
    
# MESH UTILS

class SizeInformation:
      min = None
      max = None
      size = None
      center = None
      biggest = None
      def __init__(self, min, max):
          self.min = min
          self.max = max
          self.center = self.min + ((self.max - self.min) / 2)
          self.size = self.max - self.min
          self.biggest = self.size.x
          if self.size.y > self.size.x:
              self.biggest = self.size.y
              if self.size.z > self.size.y:
                  self.biggest = self.size.z
          elif self.size.z > self.size.x:
              self.biggest = self.size.z
              
      
      def printInformation(self):
          print("Min: ", self.min)
          print("Max: ", self.max)
          print("Center: ", self.center)
          print("Size: ", self.size)
          
      def isNull(self):
          return self.size.x == 0 or self.size.y == 0 or self.size.z == 0

def getSizeAll():
    max = None
    min = None
    for item in bpy.data.objects:  
        if item.type == 'MESH':  
            for vertex in item.data.vertices:
                position = vertex.co  # + item.location
                if max == None:
                    max = copy.copy(position)
                if position.x > max.x:
                    max.x = position.x
                if position.y > max.y:
                    max.y = position.y
                if position.z > max.z:
                    max.z = position.z

                if min == None:
                    min = copy.copy(position)
                if position.x < min.x:
                    min.x = position.x
                if position.y < min.y:
                    min.y = position.y
                if position.z < min.z:
                    min.z = position.z
    if max == None:
        max = mathutils.Vector((0.0,0.0,0.0))
    if min == None:
        min = mathutils.Vector((0.0,0.0,0.0))
    return SizeInformation(min, max)

def scaleAll(scale):
    for item in bpy.data.objects:  
        if item.type == 'MESH':
           item.scale *= scale
           item.location *= scale
           bpy.context.scene.objects.active = item
           bpy.ops.object.transform_apply(scale=True)
           bpy.ops.object.transform_apply(location=True)
           bpy.ops.object.transform_apply(rotation=True)

def clean(obj,normals=False):
    bpy.ops.object.select_all(action='DESELECT')
    obj.select = True
    bpy.context.scene.objects.active = obj
    bpy.ops.object.transform_apply(scale=True)
    bpy.ops.object.transform_apply(location=True)
    bpy.ops.object.transform_apply(rotation=True)
    bpy.ops.object.mode_set(mode='EDIT')
    bpy.ops.mesh.remove_doubles()
    bpy.ops.mesh.select_all(action='SELECT')
    bpy.ops.mesh.quads_convert_to_tris()
    if normals:
        bpy.ops.mesh.normals_make_consistent(inside=False)
    bpy.ops.object.mode_set(mode='OBJECT')

# LIB END


svg1 = getArg(0)
svg2 = getArg(1)
y = float(getArg(2))
out = getArg(3)

height = 10.0
handleHeight = 1.0


bpy.ops.import_curve.svg(filepath=svg1)
bpy.ops.import_curve.svg(filepath=svg2)


for obj in getAllObjects():
    deselectAllObjects()
    setSelectedObject(obj)
    obj.select = True
    bpy.ops.object.convert()


selectAllObjects()


size = getSizeAll()

scale = y / size.size.y
scaleAll(scale)

objects = getMeshObjects()

def extrude(obj,height):
    enterObjectMode()
    deselectAllObjects()
    setSelectedObject(obj)
    enterEditMode()
    bpy.context.tool_settings.mesh_select_mode = (False, False, True)
    selectMeshAll()
    vector = mathutils.Vector((0.0,0.0,height))
    bpy.ops.mesh.extrude_region_move(MESH_OT_extrude_region={"mirror":False}, 
                                TRANSFORM_OT_translate={"value":vector,"release_confirm": False})
    
def run():
    i = 0
    for obj in objects:
        theHeight = handleHeight
        if (i < len(objects)/2):
            theHeight = height
        extrude(obj,theHeight)
        enterObjectMode()
        clean(obj)
        
        i += 1
    for obj in getMeshObjects():
        obj.select = True
run()
bpy.ops.object.select_all(action='SELECT')
bpy.ops.export_mesh.stl(filepath=out)

The end

The next post will be about a 3D people scanner machine that I’ve built šŸ˜… . Bigger failure than this above, but also more interesting.

Did you enjoy the post? Or maybe did you have a failed project like this in the past? Let me know in the comments.

This is a story of making an app called Riskometer. The aim of the app was to prevent coronavirus to spread. I will explain how the idea worked from the programmer point of view. Finally the app didnā€™t start off, because the polish government (which we were talking with for about 3 weeks) refused it, Iā€™ll explain why in the later part.

 

In march 2020, in the time of the biggest lockdown, one of my colleges asked me to join the team that is working on the app for preventing coronavirus. In the team there was already about 30 people, but they didnā€™t have any programmers. I joined, cause I like interesting projects and I thought what bad can possibly happen. Maybe our app will save someoneā€™s life.. That was worth the try.

 

The idea was simple .. in theory. Google is saving history of location of your mobile device if you have Google account connected – most people have. Now the app should take this data, and analyze it, comparing with location history of people that had covid-19 diagnosed. First, my job was to check out if it is even possible to do with the data that is provided by Google Takeout json files – the only way to get this location history. The big json file in the Takeout contains only GPS coordinates, timestamps and type of transport. The problem with this data was, the GPS coordinates are not accurate here. It can be checked if someoneā€™s path crossed the path with someone infected, but with very bad accuracy – they can be on other side of a street for example. Also checking that someone crossed the path with the infected one was not the point at all. From the government, weā€™ve get restrictions, when we can assume that someone had a contact with COVID-19 infected person:

 

  • a person lives in the same house with someone infected by COVID-19
  • a person had direct physical contact with someone infected by COVID-19
  • a person had contact with someone infected by COVID-19 in less than two meters and more than 15 minutes 
  • a healthcare worker or another person that looks after someone infected by COVID-19
  • a person that was in the same plane in the distance of two places (each direction) with someone infected by COVID-19

 

Weā€™ve focused on the point when you meet infected person for 15 minutes in less than 2 meters. That was not possible to check it just with the GPS coordinates from Google Takeout. Fortunately in the zip package, there was more files that we could use. There was a folder named ā€œSemantic Location Historyā€ that contained JSON files for each month with a bit different data. There was more useful data including activities and visited places in the sequence of visited place -> activity -> visited place -> activity..

For example if you were in some place (house, shop, mall) or you’ve travelled from place A to place B indicating which type of transport did you use like car, bus, on foot etc. 

The most accurate result that we could get from the data that worked with the rule “2 meters for 15 minutes” was to detect if someone travelled in a car with infected one. That was very easy to check with this data. There is information about starting place, ending place, starting time, ending time and type of transport. If this parameters equals to the data of another person, that is very highly probable, that those two people was travelling in the same car, so the distance between them must have been less than two meters. 

Other options were to check if someone was working in the same place, if someone live in the same house, if someone was in a shop / barber / dentist etc. where worked someone infected. But they were not as accurate as travelling in the car, so it ended just as additional information in the app that can be shown.

 

In the first few days of the project we’ve asked volunteers on our Facebook group to get 100 Google Takeout json files (from 100 different people), for the demo. We’ve get them in 3 days, so we could start with creating the mvp. The app was a very simple php (symfony + propel) application that analyzed the json files and put them into the database. After you uploaded your Google Takeout zip file, It just checked if you travelled with someone infected (randomly at first) in a car, or have you been in same place and then displaying you a list of these activities.

 

The results in this mvp looked like this:

 

 

The main problem with the Google Takeout zip packages was providing them by the user. To download the package, user has to login into his Google account, navigate to takeout.google.com, from the list of available data select only location history and then click on export. Then he waits until the package is ready for download (may be a few minutes) and then finally he can download the takeout file. After the file is downloaded, he can upload it to the app, or extract the zip package and select json files from only last month. 

 

The process was too complicated for a regular person, we had to automate it a bit.  From Google there was no any API, or any other way to get this data. There was no way to automate it using just a web application. 

 

With another programmer who recently joined the team working on bluetooth-based module for the application, we came out with an idea of using WebView component to click-through the whole download process. The only thing that user had to do was to login into his Google Account, with all the steps running automatically in the background. Two of us (with some help in design and styling from other people), weā€™ve made a working mobile app in just one weekend. It worked as it was expected, the user opened the app, logged into his Google account, and after clicking the start button all the magic happened. In the hidden WebView component, the injected javascript code clicked on each button to finally download the file. After the app detected that the file is downloaded, it unpacked the zip and uploaded required json files to the server and displaying the result.

 

That was how the final version of the app looked like:

 

 

 

 

After presenting the app to the government they asked us how much it will be cost for servers for maintaining the server side code. That was really hard to guess. With a help of another programmer that is a specialist in AWS, weā€™ve came out with the maximum monthly price of ā‚¬50k (for a whole nation using it, about 0.2 cents for person a month). After this the whole situation all started to be rough. Now the method of gaining the data from users was the problem (GDPR), there was problem that Google may be not able to maintain all the exports at the same time. Even some professor related to the government send an email to everyone involved, that the whole idea of the app is completely wrong, and he presented his research that the GPS coordinates are not accurate, because it canā€™t be compared to his professional gps. He didnā€™t even analyze the way of our app worked, but it didn’t matter. After that the whole project was down. The government said that they will pick some other project, and they did. Theyā€™ve chosen a company called Polidea that was making an open-source app called ProteGo Safe