Taylor (YC S23)’s Post

This is huge for people using language models in production—by using our classification API to identify user intent, you can route different requests to different systems. - Recommendation & Recent Events ➡️ route to search engine with up-to-date information - Easy factoid questions & chats ➡️ cheap, fast language model - Challenging questions, coding, & reasoning ➡️ powerful, capable language model Excited to see what people build with this!

Founders, your proof-of-concept should test the core capabilities your product will need to have in a real-life environment. Recently, I helped a client release the project that they defined as a proof-of-concept that was 95% ready for public release. There were lots of features and various components in the app—lots of work was done by the moment I joined. There were slight problems, though. The concept was tested on 200 records that only had limited search functionality. In real life, it had to work with 20 million records available for a full-text search with tens of facets. The proof-of-concept retrieved 200 records from a third-party API. The real-life one needed to retrieve 20 million records every month with a quota limit of 60 API requests per minute (yes, we needed to come up with some creative solutions to what seemed like an unresolvable problem). Needless to say, what was considered 95% ready resulted in around 100 hours of additional work, causing financial losses and a significant delay in release for the client. This was especially painful, as we had to rewrite some of the previously implemented features to use the new mechanisms. So, I would highly recommend focusing the proof-of-concept on the core capabilities and the real-life volumes of data it will need to work with, not features.

Have you ever wondered how software components communicate with each other? It's pretty crucial to have maintainable and flexible systems! Two common behavioral design patterns that enable effective component communication are the observer pattern and the publisher/subscriber pattern. In the observer pattern, an object (the subject) maintains a list of dependent objects (observers) and notifies them automatically when its state changes. This pattern promotes loose coupling by eliminating the need for observers to constantly check the subject's state. For example, a spreadsheet application may use the observer pattern so charts observing a dataset get updated when data changes. The publisher/subscriber pattern builds on the observer pattern, but uses an event system allowing publishers to broadcast events without knowing what subscribers there are. Subscribers can listen for events of interest without tight coupling to publishers. For example, a messaging app could let chat rooms publish chat events subscribers join to receive. Publisher -> Topic / Event Channel <-> Subscriber Using patterns like observer and publisher/subscriber leads to flexible, modular code. Components can communicate without tight coupling or reliance on each other's implementations. This makes it easier to maintain and extend systems over time. The next time you're designing interactions between components, consider leveraging one of these behavioral patterns! #softwaredesign #designpatterns #observerpattern #publishsubscribe #eventdriven #loosecoupling #softwarearchitecture #modularcode #softwaredevelopment

Technology update: What have I learnt from building User Interfaces? https://lnkd.in/ekKbyc5Z

Looking to automate tasks? What software can help you with web scraping, idea checking, and information gathering for free?   #AutomateYourLife #InnovativeSolutions #SoftwareIdeas Hey everyone! 👋 Do you need something to automate? Whether it's for your work or personal life, I want to hear from you! Here's the deal: I'm looking to build a product or service that can help solve everyday problems, but I need your input. What software or tools do you wish you had to make your life easier? Here's how you can contribute:... Source: https://lnkd.in/gMAFsGgS #mymetric360

Looking to automate tasks? What software can help you with web scraping, idea checking, and information gathering for free?   #AutomateYourLife #InnovativeSolutions #SoftwareIdeas Hey everyone! 👋 Do you need something to automate? Whether it's for your work or personal life, I want to hear from you! Here's the deal: I'm looking to build a product or service that can help solve everyday problems, but I need your input. What software or tools do you wish you had to make your life easier? Here's how you can contribute:... Source: https://lnkd.in/gJQ6rmEn #mymetric360

Here's an example of how to use aider to do a small feature implementation: First I ask it to take an extra arg and filter. files: contents.py get_youtube_transcripts.py                                          > refactor get_youtube_transcripts to take an optional influencer id, and filter the content based on that Then I asked it to add a test case. files: contents.py get_youtube_transcripts.py tests/dream100_tests/https://lnkd.in/gMdHkB_W                > Add a test case to `test_get_youtube_transcripts` that accomodates for passing a single influencer_id The test case failed, because it did the setups incorrectly. So I had to add my conftest.py so it had the proper context to set up the test. files: contents.py get_youtube_transcripts.py conftest.py https://lnkd.in/gMdHkB_W       > refactor the setup in get_youtube_transcripts_with_influencer_id to make use of the fixtures in conftest and create an influencer and web properly corre ctly for the create_content call After this call, the test passed, and I can move on to the next feature. It's not exactly insert quarter and get application out, is it?? But, it's better than doing that refactor by hand. This is not a tool for the lazy or the uneducated, but if you're disciplined and know what you're doing, it can really take a lot of work out of things. If you're interested in learning more about how to build software with LLM's, follow me.

Sharing some general steps that how we can make our app bug free and counter corner cases. Here are some steps: 1. Identify potential corner cases: Analyze your application and identify scenarios that may lead to corner cases or unexpected behavior. This could include handling edge inputs, invalid data, asynchronous operations, or exceptional conditions. 2. Write test cases: Create test cases specifically targeting the identified corner cases. Use a testing framework to write unit tests or integration tests that cover these scenarios. Test both the expected behavior and any error conditions that may arise. 3. Handle input validation: Implement input validation mechanisms to prevent or handle invalid data or user inputs. Use techniques like form validation, data type checking, and sanitization to ensure that your application can handle unexpected inputs gracefully. Currently, we are using react hook forms 4. Error handling and fallbacks: Implement error handling mechanisms throughout your application. Use techniques like error boundaries, try-catch blocks, or error-handling middleware to catch and handle errors. Provide meaningful error messages or fallback UI components to inform users about the encountered issues. 5. Graceful degradation: Plan for scenarios where certain features or dependencies may not be available or may fail. Implement fallback mechanisms or alternative approaches to ensure that your application can still function or provide a degraded but usable experience. This could involve using default values, alternative APIs, or alternative rendering components. 6. Test with real-world scenarios: Test your application with realistic data and usage patterns to identify and address any corner cases that may arise during actual usage. This can help uncover edge cases that may not have been considered during development. 7. Monitor and gather feedback: Continuously monitor your application in production and gather feedback from users. Keep an eye out for any unexpected behavior or corner cases that arise in real-world usage. Use tools like error tracking, analytics, or user feedback mechanisms to gather insights and address any issues that arise.

Looking to automate tasks? What software can help you with web scraping, idea checking, and information gathering for free?   #AutomateYourLife #InnovativeSolutions #SoftwareIdeas Hey everyone! 👋 Do you need something to automate? Whether it's for your work or personal life, I want to hear from you! Here's the deal: I'm looking to build a product or service that can help solve everyday problems, but I need your input. What software or tools do you wish you had to make your life easier? Here's how you can contribute:... Source: https://lnkd.in/g-hRyfrh #mymetric360

Simulating users behaviors on your API ? I've been building APIs for a while but often, I find myself looking for tools I could use to simulate user behaviors. eg. signup -> login -> get feed -> create post This can be a users activity which call several APIs. Is there a tool that can do this ? + you can configure the user flow eg. signup -> login -> get feed -> comment + provide random data required for the different endpoints + use info returned from previous api call to call a new endpoint + simulate a number of virtual users all going through the user flow Does anyone find a need for this kind of tool and what would be use cases ?