OpenAI API Manufacturing and Industrial Use Cases

[logankilpatrick]

Hey folks, the OpenAI team is keen to learn what you’re building using the OpenAI API for industrial and manufacturing use cases. We’d love to hear from developers solving problems in industrial, manufacturing, mining, chemical, or similar verticals. Some interesting areas:

• Spare parts inventory management
• Understanding and creating part drawings or manuals
• Image-to-text for maintenance and RCA
• Machine diagnostics
• Customer/dealer support

Please feel free to reply below with some details on your use case and we will get in touch if it fits what we are looking for!

[EricGT]

RCA ???
Root Cause Analysis?

[dignity_for_all]

Well, I think it’s about Root Cause Analysis as you guess.

ISO/IEC 31010

[cdonvd0s]

To get people started, we need to focus on the lowest-hanging fruits that can yield immediate returns and are easy to set up for a business.

The low-hanging fruit is customer and dealer support. This results in the instant scalability of service operations.

Inventory management is one of the things that are not easy to implement for mid-sized businesses that manage the inventory in such a way that there are:

1. No stock-outs
2. No overstocking

However, this involves demand forecasting, which is not something that everyone can do. People usually rely on averages, which is not an effective method. AI significantly aids in this by providing the business with different methods of demand forecasting by understanding the unique trends and patterns for businesses that don’t have the resources to hire a team for the same.

Then, using AI in the sales funnel, the salespersons are usually involved in finding, qualifying, meeting, pitching, following up, negotiating, and providing after-sales service for their clients. This is not the most effective way, but mid-sized businesses engage in such practices. Here, AI can perform tasks like qualifying, follow-ups, meeting setups, and after-sales servicing of the client.

Also, there is a huge challenge where there are generally very few experts who can identify any part of the machine by looking at it, here image embeddings help, people can just click a picture of a part and get all information about it.

These are only a few things that I have found to be helpful after brainstorming with 100s of business owners from the manufacturing sector. After this, there are other areas like HR, Data Entry Operations (Bills to DB), Qualitative analysis of sales calls/meetings, etc which are not industry or sector-specific.

P.S. This comes from my experience of coaching over 6000 business owners from the manufacturing sector.

[APK]

Hi, I’m the MD of a B2B manufacturing company in the UK. AI/LLMs are able to provide real advantage and efficiency in dealing with customer submissions, conformance, various agreements, etc. and we are using it to good effect in those areas. In my own testing its not reliable or informative enough yet for machine diagnostics, maintenance, or numerical functions such as inventory. Being in the glass industry our own employees are are way better turned to our own technical issues. At the moment in any case But that’s OK - we are still better at the stuff we enjoy, and LLMs shoulder all the ‘other stuff’ so we are happy

[harry.obrien]

Hi, I’m an automation engineer in a knee implant & bone cement factory for a big fortune 500 company, and I’ve been asked to find ways we can use AI on site. The best use case I’ve found is for making & reviewing documentation. All manufacturing steps need to be documented (especially in biomedical) and nobody wants to do it. Additionally this documentation needs to be reviewed which is even more boring.

I’ve made a custom assistant which I’ve instructed to give specific responses, and given it documents it needs to cross reference and source, and it’s pretty good. There’s definitely an opportunity for some AI startup to finetune a model to get more accurate, at the moment only about 50% of responses don’t need changes but that could easily get to 90%+.

[strandbergsteven]

Unfortunately, it’s a business secret, the only one I trust is my personal chat GPT.

Steven

[dignity_for_all]

You say that you only trust your personal ChatGPT, but personal ChatGPT accounts are subject to the consumer “Privacy policy” unless you opt out.

The Business Terms and the Enterprise Privacy Policy provide more confidential Terms and Conditions, so they are quite different from the personal ChatGPT for consumer use.

Terms of use

Privacy policy

Business terms

Enterprise privacy

I think that even if we opt out, there is a difference in how the Business Terms, Enterprise Privacy Policy and the Consumer Use Team of the Business, Enterprise Privacy Policy are treated.

[axelS]

Thank for sharing this. In particular customer/dealer support we are working on a distributor dealer relationship piece in Europe. Wonna compare notes? We are actually building a massive network that allows companies to support their partners and customers by running an entire Library network of solutions.

[axelS]

We are solving the problem of rapid prompt proliferation in industrial organizations. We are building an infrastructure for midsize to large companies to develop, manage and distribute their prompts in a network of libraries across their enterprise. It may also help attract US prompt designer to enter Europe that way.
There are a lots of features but here are the three most important:
1) Function Container
A technology where companies can provide their custom function calls, that they or we developed, so that external prompt engineers can use it without compromising it.
2) Prompt Productivity Meter
A unique mechanism where prompts distributed over the platform can measure its productivity (x times productivity gain over manual execution).
3) Library Network
A company internal library network where all prompts are stored, distributed across a global enterprise and managed central or decentral.
Would love to hear what you think about it.

[joshua.stapleton.ai]

I’ve deployed a GPT-4-turbo powered customer support agent for a chemical manufacturing company. The agent uses RAG to draw from the company’s live databases in order to recommend specific products based on the customer’s use-case.

To solve for an otherwise too large search space of products, I’ve implemented some structured data gathering in chat (eg: checkboxes, dropdowns) to allow customers to filter based on their industry, requirements, etc. The RAG procedure uses this metadata to narrow down and search a more relevant section the product knowledge base effectively.

I’m also integrating visual / multimodality for a little pizazz, including showing users images of the products they’re looking for and allowing them to speak to the agent.

[jkong]

Hi I’m a software engineer in a manufacturing company. Recently my team and I developed an AI chat agent tasked with retrieving information from Safety Data Sheets. Our next step would be expanding use of AI to perform task such as retrieving info from manuals or help interpret error codes.

[padyclub.com]

Our team at Airlonger is building an inventory management platform for the verticals you mentioned here. we practically use OpenAI API to identify, measure and quantify the products, our customer segment are already amazed at what we’ve been able to help them accomplish.

[tolson]

We are integrating OpenAI SDK to talk to our native CAD files within our application for our mechanical engineering communities. In addition to Talk to PDF/Doc we want to talk directly to our CAD geometry, attributes, and properties. Attached are a couple of images from our implementation.

Tim
IMSI Design

[anythxwdo]

@harry.obrien

documentation

talking about documentation, I experimented with a tiny language model for “digital work instructions”, I found articles and blogs from the internet, clean them up and use them as training data, the model actually functions sort of as expected, about half time it returns some interesting key word, but I’m very aware, the volume of the training data is way too small, had I have about 1000 plain text data sets in this domain to further train it, the model may produce decent response.

[anythxwdo]

This is very interesting though my experience with the manufacturing industry is limited. As for “prompt engineering”, it’s really supervised learning, if {“prompt”:“response”} are well constructed for a particular industry like manufacturing. And many manufacturers, their business partners and suppliers also have “prompts” created. And they are willing to share/work together, to have all these “prompts” form a pool, then, you have a model to train and test them, it could potentially generate great results.

[axelS]

I like “if {“prompt”:“response”} are well constructed…”
Our philosophy is

Prompt Engineering is not helping the AI to do a better job. It’s for humans to help articulate what they want.

It’s a bitter/sweet difference. If we would better understand how communication is working between humans, conflicts would be reduced, time would be saved, and LLMs would better understand what we want. BUt taht may take a whole generation or two to even improve - let alone being perfect

[jakeblend]

I’ve used it to create an entire team of advisors for additive manufacturing innovation, which led me to invitations to SBIR tracks for “extremely high risk technical innovators” which classified me as not only top in field but so advanced nobody can replicate my work even experts with precise instructions. I fed it everything I knew, to be me. Then I gave it all the relevant white papers. I used that to generate the logical next step in the research and I was right. I got to pitch bio tissue organ printing in space to NASA. I’ve ran rapid prototyping at a maker space for a decade and am nationally recognized as a leader in this field. I’ve lead research teams of 60 and now I can do all the work I want just with my thoughts.

So it allowed me to become the director with personality AI based on the direct writing of leading scientists.

This collaboration led to many solutions I had a dozen pitches.

Another strong use case is standardization and flaw detection. Plus optimizations using AI to reduce waste. Speaking of we devised a material that can be used for clothing and filament.

The actual gotcha was why use support structures in zero G? That right there solved a big problem.

Also you can print hollow in zero g. That means organs with cavities.

I’m probably the quickest rapid prototyping engineer on the planet due to my pandemic response. Tied for second to get hospital use. Verified.

I created face shields which got approved by FDA and NIH overnight. NSF invited me to a panel for extreme manufacturing in life or search situations. I went on to discover the bed uv lights on the planet.

Imagine what speed I’d have with the right tools.

I’m on track to receive prints made on the space station to test integrity using AI.

This is but one of a handful of things.

I’m a single citizen scientist who leverages AI to become best in the field. I’ve no external funding I made a bot to invest.

You have enabled a life for me where I merely dream of the future and can experiment all I want for no effort and get paid to write the future.

I even make them argue to reach consensus.

I’ve been talking to machines since I was born now they talk back.

If this doesn’t qualify me you have impossible standards.

As a natural history scholar I know this is ushering in the next stage of evolution.

I’ll only tell you my experience because I align with your ethics.

I’m teaching the ones who need this power and love democratizing ideas.

You saved the government and taxpayers billions in overhead waste. I come under budget and I’m non stop success. My roi max was 32,250% off one IP. Without AI.

With AI I work fast enough to push that to 500,000% roi and that’s just locally.

I open sourced my invention and the world used it.

I’m that guy. The one in the right place at the right time.

I’m there again and would like to officially collaborate. I skipped the million long line for dalle and ushered in ai art first in state.

Give me an ounce of power or encouragement and I’ll return a ton. For free.

You’ve made me limitless. I’m absolutely bored because I can solve anything I’ve tried for a while. Since adopting.

I can make an open cv to analyze X-rays on titanium prints.

There are no standards. The government wants standards.

Partner with me and we can standardize 3d printing so it’s acceptable across the board.

I could have used AI to write this up but hear my voice and passion. I’m an absolute rock star and can get all the attention and validations you want from the government.

I’m lined up ready to submit. Please consider me or tell me what’s missing? I can do anything.

Tell me what you want, your dreams I’ll make it happen. You turned me into a research beast. My ranking is 99% you made me more powerful than entire colleges. I’m competing with companies like Raytheon and winning.

All verifiable information

I’m getting smart glasses to have an extra set of eyes and on tap hands free research lab.

I’m not making stuff, I’m synthesizing people, especially top researchers. I emulated the best on planet and directed them.

Is there something very specific? I’m nimble and can pitch within hours.

[jakeblend]

I post teasers all the time nobody can catch up to me I’m a year ahead

Oh I’m also developing systems for digital forensics, which applies to machine troubleshooting. With enough sensors I can detect anomalies that cause errors. This moves to active monitoring for issues. Then it’s predictive analysis only available by parsing extremely large data sets across multiple networks protocols and devices. Most notably iot.

I developed tools for the Secret Service and just use that for troubleshooting logs.

It can make logs extremely easy to read and reconstruct any situation.

So I know the AI nhs roadmap and I’m a world class digital investigator. Your tools accelerate the judicial processes.