Seeking Honest Feedback: Review MVP NoCode ML
Mar 4, 2024
Thank you for all your comments and feedback on my previous post.
Taking your insights into account, I've developed an MVP for a no-code machine learning tool tailored specifically for tabular data, and I'd greatly appreciate your input.
https://nocode-machine-learning.streamlit.app/
Currently, the web app can clean up data, offer tips for excluding incomplete columns, and analyze it using the random forest algorithm to detect significant trends.
The aim of this tool is to provide non-technical users with a simple solution for exploring data trends without needing coding skills or machine learning knowledge.
While I acknowledge that the web app is still in its early stages and offers limited functionality, I'm eager to gather insights on the following:
a) Is such a tool useful?
b) Identify which features are most beneficial
c) Learn about similar solutions already in the market
Thanks for taking the time to read this, and I look forward to hearing from you!
Feb 20, 2024
Hey everyone,
I'm developing a machine learning web app aimed at professionals outside the tech sphere. It's a simple tool where you can input your data, and the web app will clean it up, provide some generic tips of excluding incomplete columns, analyze it using random forest algorithm and try to identify if any significant trends exist and if a more detailed ML model is worth considering.
The idea behind this tool is to bridge the gap between data and non-machine learning professionals. Whether you're in marketing, finance, or healthcare, if you work with data but lack expertise in ML, I'm hoping such a tool is useful. But then again, I could be wrong.
Do you think such a tool is needed? Have you come across anything similar? Any issues I'm overlooking e.g. I realise ML is an art in itself, but I wonder if there is a simplistic good enough approach that can help guide what projects might be worth investing in. Would you find this helpful in your work? Any thoughts, comments, or suggestions would be immensely valuable.
Thanks for taking the time to read this, and I'm looking forward to hearing your honest opinions.
Best,
There's two ways to go about it and potentially you can combine it.
1st, skip the ML & AI language and just focus on the use cases - data trend forecaster or whatever.
Then from there you can decide to educate them on ML or just give them the tools without even having to let them know how the "food is made" basically.
Even with the first approach, I'd consider having a separate user persona of nontechies looking to get into ML and have a totally different approach for them marketing wise. At minimum you'd need a separate landing page, some SEO articles about "ML for nontechies" and I would recommend a course/ newsletter on the subject.
Thanks for this @aarraudits - this is very insightful. What do you think about the MVP I just created?
I'm trying to focus on the use cases at the moment, and instead of educating them on ML, I'm just stating the steps taken for data cleanup. Is that adding value or confusing things?
What do you think the user persona would look like? Folks interesting in learning ML or industry people just interested in trend or no trend analysis to determine if further time/money investment is worth it on the given dataset?
No
@drshivsidana Could you elaborate further? Do you think the answer is no because other similar tools exist that are doing the job, or are you suggesting that there is no need for such a tool or further are you of the opinion that the topic of ML is so complex that it is practically impossible to create an automated tool that could provide useful results without human analysis?
Tackle the use case first. A lot of non tech folks love the AI/ML innovation around and want to dive into it, however they have no clue how and WHAT they should solve with it. Maybe you should look at some of their problems and pick up niche for no code ml tool.
When you say niche do you mean in terms of technology i.e. random forest vs linear regression or do you mean niche in terms of a ML pre-processing step of data clean up etc?
Any ideas on how to determine my use case (except myself :P) - are there platforms like indiehacker that I can determine/test use cases?
Thanks @aminmemon for the feedback, appreciate it.
Setting aside quality control issues for arbitrary user-submitted input data, you’re going to want to put way more focus than you think you need to on messaging and creating case studies. Getting non-ML natives to trust blackbox-y ML approaches to their problems when they’re used to traditional workflows or business logic sets is a massive hurdle. (I work for a company applying ML in healthcare space.) Oh yeah, and getting them to trust your startup with their data will be its own issue. Security infra and messaging around that also critical.
Very interesting insights @duffdevone - I was hoping to use streamlit and keep computing small to begin with so I don't need to use 3rd party GPU's etc. Currently, all my web apps have 'we don't store your data' within the disclaimer but I suppose you think much more explicit messaging will be needed to ensure data security.
Any ideas of how/where I can get in touch with non-ML native potential users/companies to understand their fears?
Thanks for the feedback, very well appreciated.
The disclaimer is a great start for sure. Am probably biased coming from world of healthcare and finance. No one wants to trust you with their data, even in transit 😂
As it relates to lead gen, I understand the goal is to have a generalist app. That said I wonder if you’d have more successs finding businesses to talk to if you narrowed the scope of who this is for. Niche down to solve a specific industry problem (eg predicting user behavior in retail ecom), then expand. Also instead of purely self serve for end user, could even offer to do some consulting for free/cheap to get some case studies.
Random detail.. if you haven’t, might want to give some consideration to feature importance. Biz will likely want to know the main drivers of model performance.
Sounds exciting! Definitely a huge need out there.
@duffdevone absolutely great insights. Based on your comments, I've added a feature importance summary after the RandomForest analysis is optimized in my MVP listed above. Let me know what you think? Any further insights?
I've also tried to kee the scope limited at the moment to tabular numeric or categorical data for numeric data-intensive industries like engineering etc. I'm not sure what the scope of consulting services could be, should I not focus on creating a web app that is self-sufficient? Unless you think that ML is too complex for this to be feasible?
Nice — feature importance makes the output more meaningful.
I definitely agree that full self-service should be the aspiration. Was more wondering if you might need to evangelize/be more hands on with clients early to build up some case studies.
Honestly though you could just create some case studies/working examples on your own. Would kill two birds with one stone:
Thanks for such meaningful insights! Tutorials and practical demonstrations are great ideas I can work on quickly too.
Let me know if I can assist you in your projects in any way :) , would love to be of value.
Interesting concept! Consider having clear disclaimers and guidance
That makes sense @premsaini - I will look into that to ensure the limitations of the product are explicit. Thanks for that.
Who is the user? What is their current workflow? Where does your service fit in, in what they currently do? Are they the one's paying the bill or would you need to also sell it to a 2nd stakeholder that carries the budget?
Is it for business analysts? What are their current data sources and tools? Do you integrate with said tools and how or do they need to make that up themselves/manually pass the data to your service and back?...
How do you establish trust about their data? What certs/guarantees would you offer? Would you be able to comply with data locality concerns?
Or can you find some bigger tooling framework that this can be offered inside where trust, tooling and workflow are already established and you'd be a paid extension/add-on..
Very interesting questions hatkyinc.
My initial thought was that this would be for people detached from the ML world e.g. mechanical engineers in oil and gas etc who have large sets of data and may want to see if ML can add accuracy or efficiency but do not have the funding or conviction yet.
I was going to use streamlit and keep the processing to smaller data sets where a GPU is not needed and therefore have no overhead processing cost, analyses the data locally and present charts of the analysis within the web app itself. I'm assuming this would address the data storing concern as I understand it does not get stored anywhere after the webapp is closed.
Really like the idea of this fitting inside a larger framework/tool, any ideas of what kind of tools would it add value to?
Thanks a ton for your thoughtful questions, I'm going to research and explore these further as I'm new to the tech world myself.
IDK these niches to much, you probably need to interact with some people in the niches your targeting, what are the job titles? In big companies you could have many players in the middle and some kind of parallel paths of the data, some going to data warehouse with ETL tools and people responsibility for the data paths and transformation, some people that pre process the data into business concepts and some different groups that try to query the data, some more technical and several more that are end customers with lower skills, but these people might not have budgets/privileges/authority to add tools
Firstly you need to identify your target customers and it's limitations, interactions ect
ETL tools have flows that might be pluggable but I don't have much details into it, data flow tools, not my niche
@hatkyinc I agree with this, unfortunately I'm not sure how to handle the middle players. I was hoping to create a tool to show value to the middle players and the decision-makers though I doubt that will be enough. Any ideas?