Endless (https://www.endless.academy) will build a course for you, on any topic. You can customize what's taught and the teaching style. It's an efficient way to learn something enough to start using it.
There are a LOT of projects like this. Endless is different in important ways:
- It's block-based, like Notion. You can use this to customize the teaching style, by adding video, exercise, or other blocks.\n - Courses are dynamic. You can add new blocks to an existing course, and even edit the course.\n - AI chat blocks let you get help. I think chat should be integrated into the flow of the course, not separated into a different window. This lets you have multiple chats going, and preserve the context in each one.\n - Steer how the AI develops the course by editing the outline.\n - Endless looks up information from high-quality educational resources (you can see the sources)\n - It uses that information to prompt an LLM to write the learning blocks\n - If necessary, it uses other models and services to post-process each block (for example, to find a video)\n - I wanted the interface to feel dynamic, with information from the LLM streaming in realtime. This meant I had to rip everything apart halfway through and make the backend mostly async (a paradigm I hadn't used before).\n - SvelteKit is fantastic - everything being plain Javascript/HTML makes creating components fast. I only ran into a couple of issues where the magic made it hard to understand errors. That said, there are so many off the shelf React components that overall productivity may not have been that different.\n - I used TipTap for the block editor, and I mostly like it. My main issue is handling state across multiple TipTap instances interspersed with Svelte components. The "correct" way seems to be using nodeviews and a single TipTap instance, but that trades off data model flexibility (since TipTap imposes its own schema).\n - I did not use Langchain/LlamaIndex/etc. In production, the need for reliability and observability meant that I just wrote my own light wrappers instead.Endless (https://www.endless.academy) will build a course for you, on any topic. You can customize what's taught and the teaching style. It's an efficient way to learn something enough to start using it.
There are a LOT of projects like this. Endless is different in important ways:
- It's block-based, like Notion. You can use this to customize the teaching style, by adding video, exercise, or other blocks.\n - Courses are dynamic. You can add new blocks to an existing course, and even edit the course.\n - AI chat blocks let you get help. I think chat should be integrated into the flow of the course, not separated into a different window. This lets you have multiple chats going, and preserve the context in each one.\n - Steer how the AI develops the course by editing the outline.\n - Endless looks up information from high-quality educational resources (you can see the sources)\n - It uses that information to prompt an LLM to write the learning blocks\n - If necessary, it uses other models and services to post-process each block (for example, to find a video)\n - I wanted the interface to feel dynamic, with information from the LLM streaming in realtime. This meant I had to rip everything apart halfway through and make the backend mostly async (a paradigm I hadn't used before).\n - SvelteKit is fantastic - everything being plain Javascript/HTML makes creating components fast. I only ran into a couple of issues where the magic made it hard to understand errors. That said, there are so many off the shelf React components that overall productivity may not have been that different.\n - I used TipTap for the block editor, and I mostly like it. My main issue is handling state across multiple TipTap instances interspersed with Svelte components. The "correct" way seems to be using nodeviews and a single TipTap instance, but that trades off data model flexibility (since TipTap imposes its own schema).\n - I did not use Langchain/LlamaIndex/etc. In production, the need for reliability and observability meant that I just wrote my own light wrappers instead.However with all the news surrounding tech lately, mainly the advancement of AI, I've become somewhat discouraged about this whole field, and as I result I haven't been putting a lot of effort into my studies, to put it lightly. I'm barely getting passing grades. Over the previous spring and summer, before switching majors, I did a Web Dev bootcamp and I really enjoyed programming with the abomination that is JavaScript, but now I don't even like programming anymore.
I will not consider going back to my old major, since it's easily replaceable by AI.
I'm thinking of getting into agriculture (I have the necessary capital) and going down that line of work/self-employment (which I vastly prefer over being a salaried employee.)
Sure it can be physically demanding work (which I don't mind), but I don't see AI getting to that anytime soon (and to the extent that it's gotten into tech as we know it), and I have done that kind of work in the past and it's just inherently more fulfilling than staring into a computer screen, to me. Plus, it pays fine-ish and it can pay very well if, down the line, I open a food processing factory.
Any advice or thoughts you have to offer?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Is there a point in me getting a CS degree anymore?"}},"_tags":["story","author_vik0","story_42859394","ask_hn"],"author":"vik0","children":[42859493,42859555,42859884,42860270,42860588,42861259,42861418,42861866,42861926,42862527,42863321,42863355,42863411,42863580,42864354,42871430,42875935,42898376],"created_at":"2025-01-28T23:11:17Z","created_at_i":1738105877,"num_comments":38,"objectID":"42859394","points":16,"story_id":42859394,"story_text":"I'm pursuing a CS degree right now. I used to do another major (I'm from Europe and had the equivalent of a little less than a 4.0 gpa), but I dropped out of there because I thought a CS major would be better.
However with all the news surrounding tech lately, mainly the advancement of AI, I've become somewhat discouraged about this whole field, and as I result I haven't been putting a lot of effort into my studies, to put it lightly. I'm barely getting passing grades. Over the previous spring and summer, before switching majors, I did a Web Dev bootcamp and I really enjoyed programming with the abomination that is JavaScript, but now I don't even like programming anymore.
I will not consider going back to my old major, since it's easily replaceable by AI.
I'm thinking of getting into agriculture (I have the necessary capital) and going down that line of work/self-employment (which I vastly prefer over being a salaried employee.)
Sure it can be physically demanding work (which I don't mind), but I don't see AI getting to that anytime soon (and to the extent that it's gotten into tech as we know it), and I have done that kind of work in the past and it's just inherently more fulfilling than staring into a computer screen, to me. Plus, it pays fine-ish and it can pay very well if, down the line, I open a food processing factory.
Any advice or thoughts you have to offer?","title":"Ask HN: Is there a point in me getting a CS degree anymore?","updated_at":"2026-01-28T04:50:44Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ellis0n"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["vikk","ai"],"value":"Hello from Ukraine,
This is Animation CPU development team and we have been building smartest the world's smartest live coding programming platform for mobile since 2012
Animation CPU made from the scratch.
Here is demo of IDE and HN API demo https://vimeo.com/manage/videos/732098541 and here is creative work https://vimeo.com/manage/videos/732128866
Screenshot: https://twitter.com/AcpuStudio/status/1550132877752279043/photo/1 and video https://twitter.com/ClubAvatarapp1/status/1550150060863901698
Build
- Metaverses
- Games
- Creative works & Clips
- Apps
- Presentations
- Sites
Language & environment features:
- realtime profiling
- realtime time travel debugger (just like Bret Victor)
- single variable type - evertthing is float (GPU & AI optimized)
- structured logs
- everything is formula, formula is live code
- everything is node - light thread or process
- simple standard library for everything
- designed for microcontrollers
- pure & secure
- no JIT
- never halt
- never crash
OS features:
- target platform independent
- custom bootloader
User features:
- Twitch streaming
- Video recording
- Networking HTTP & P2P
- Builtin code sync & live update
Development features
- built-in IDE
- designed visual programming, everything is visible
- UI/UX and no-code
- everything is linked, navigation between UI, code and vm
- everything is logging, save history with screens and video
- export static rom
Blockchain:
- Ethereum integration, see https://twitter.com/web3wallet
Team and social features:
- designed for team building
- payment voting for formulas
- chat
- platform for building and improve all levels of apps like browser plugins
AI:
- designed for AI
In order to fulfill our dreams of mobile coding from scratch, a unique architecture has been developed:
- system of protection against critical errors
- blocking long loops and recursions
- blocking memory leaks
ACPUL programming language - a minimalistic language with a low entry threshold
- very fast compilation
- modularity and flexibility
- is a level 0 assembler for the ability to create more advanced languages and tools based on it
Animation CPU is an OS capable of running on bare matal due to its modularity and minimal dependencies.
If you would like to try AnimationCPU on your Apple device, we can submit TestFlight, please email me viki.tux25@gmail.com
Due to lack of resources, our website animationcpu.com is very primitive right now, but we are working on delivery to the AppStore.
Please follow Twitter and Instgram https://twitter.com/AcpuStudio
Support us: bc1qnkj44rdj790razlt6p4js9y2ay8qwc5dntu2yu
Or just buy coffee with Paypal: viki.tux25@gmail.com"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Animation CPU programming platform for Apple"}},"_tags":["story","author_ellis0n","story_32181251","show_hn"],"author":"ellis0n","created_at":"2022-07-21T16:21:52Z","created_at_i":1658420512,"num_comments":0,"objectID":"32181251","points":2,"story_id":32181251,"story_text":"Hello from Ukraine,
This is Animation CPU development team and we have been building smartest the world's smartest live coding programming platform for mobile since 2012
Animation CPU made from the scratch.
Here is demo of IDE and HN API demo https://vimeo.com/manage/videos/732098541 and here is creative work https://vimeo.com/manage/videos/732128866
Screenshot: https://twitter.com/AcpuStudio/status/1550132877752279043/photo/1 and video https://twitter.com/ClubAvatarapp1/status/1550150060863901698
Build
- Metaverses
- Games
- Creative works & Clips
- Apps
- Presentations
- Sites
Language & environment features:
- realtime profiling
- realtime time travel debugger (just like Bret Victor)
- single variable type - evertthing is float (GPU & AI optimized)
- structured logs
- everything is formula, formula is live code
- everything is node - light thread or process
- simple standard library for everything
- designed for microcontrollers
- pure & secure
- no JIT
- never halt
- never crash
OS features:
- target platform independent
- custom bootloader
User features:
- Twitch streaming
- Video recording
- Networking HTTP & P2P
- Builtin code sync & live update
Development features
- built-in IDE
- designed visual programming, everything is visible
- UI/UX and no-code
- everything is linked, navigation between UI, code and vm
- everything is logging, save history with screens and video
- export static rom
Blockchain:
- Ethereum integration, see https://twitter.com/web3wallet
Team and social features:
- designed for team building
- payment voting for formulas
- chat
- platform for building and improve all levels of apps like browser plugins
AI:
- designed for AI
In order to fulfill our dreams of mobile coding from scratch, a unique architecture has been developed:
- system of protection against critical errors
- blocking long loops and recursions
- blocking memory leaks
ACPUL programming language - a minimalistic language with a low entry threshold
- very fast compilation
- modularity and flexibility
- is a level 0 assembler for the ability to create more advanced languages and tools based on it
Animation CPU is an OS capable of running on bare matal due to its modularity and minimal dependencies.
If you would like to try AnimationCPU on your Apple device, we can submit TestFlight, please email me viki.tux25@gmail.com
Due to lack of resources, our website animationcpu.com is very primitive right now, but we are working on delivery to the AppStore.
Please follow Twitter and Instgram https://twitter.com/AcpuStudio
Support us: bc1qnkj44rdj790razlt6p4js9y2ay8qwc5dntu2yu
Or just buy coffee with Paypal: viki.tux25@gmail.com","title":"Show HN: Animation CPU programming platform for Apple","updated_at":"2024-09-20T11:41:47Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Ansh_Dhanani"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["vikk","ai"],"value":"1000 lines of AI code. zero idea what breaks.
Here's what actually happens when you vibe code a real project:
You ask your AI to add a feature. It writes the code. Looks right. You ship it. Then something breaks three layers deep \u2014 a function that imported from the file it just rewrote, a module boundary it didn't know existed, a constraint your team defined six months ago. Your AI had no idea. It never does.
The problem isn't the model. The problem is that every AI coding tool gives your AI a flat list of files and asks it to infer architecture from that. It can't. It guesses. And on non-trivial codebases, the guesses compound.
Specifically:\n- it doesn't know which functions depend on the file it's editing\n- it doesn't know your module ownership or layer rules\n- it doesn't know what changed since the last prompt\n- it has no way to validate its plan against your actual architecture before writing
Mikk fixes this by building a directed graph \u2014 the Mesh \u2014 of your entire codebase. Not a summary. Not embeddings. An actual graph: every function as a node, every import and call as a typed edge, forward and reverse adjacency maps for O(1) traversal in either direction.
`mikk init` runs in ~3 seconds on a 90-file project:\n- 2,847 functions parsed via TypeScript Compiler API (real AST, not regex)\n- 9,442 edges in the dependency graph\n- SHA-256 Merkle tree at function \u2192 file \u2192 module \u2192 root level\n- everything written to mikk.lock.json, ~60% smaller than raw source\n- claude.md + AGENTS.md generated with tiered architecture context
Then an MCP server exposes the graph. Your AI doesn't get a paste. It calls tools against the real graph mid-task.
Three things that actually matter:
1. blast radius before the edit\n`mikk_before_edit` walks the reverse dependency map from the file you're about to touch. Returns every upstream caller, every exported API at risk, every constraint that applies. Your AI sees the full blast radius before it writes a single line. `mikk_impact_analysis` goes deeper \u2014 transitive blast radius across the entire graph, estimated impact level, public API consumers.
2. context that's not RAG\nThe AI Context Builder seeds from your task description, BFS-traverses the call graph outward, scores each reachable function by proximity + keyword match + entry-point bonus, then greedily packs the highest-scoring functions into your token budget. Every result includes exact file path, start line, end line, and full source body. No hallucinated paths. No wrong modules.
3. architecture that enforces itself\nDefine module boundaries in mikk.json. Six constraint types: no-import, must-use, no-call, layer, naming, max-files. `mikk contract validate --strict` fails CI with exact file + rule citations. ADRs sit alongside constraints so your AI knows why a rule exists, not just that it does.
Local-first. No cloud. No telemetry. No API keys. TypeScript, JavaScript, Go.
https://github.com/ansh-dhanani/mikk\nhttps://mikk-web.vercel.app"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["vikk","ai"],"value":"Mikk \u2013 your AI wrote the code but doesn't know what breaks"}},"_tags":["story","author_Ansh_Dhanani","story_47387807","ask_hn"],"author":"Ansh_Dhanani","created_at":"2026-03-15T14:38:08Z","created_at_i":1773585488,"num_comments":0,"objectID":"47387807","points":1,"story_id":47387807,"story_text":"1000 lines of AI code. zero idea what breaks.
Here's what actually happens when you vibe code a real project:
You ask your AI to add a feature. It writes the code. Looks right. You ship it. Then something breaks three layers deep \u2014 a function that imported from the file it just rewrote, a module boundary it didn't know existed, a constraint your team defined six months ago. Your AI had no idea. It never does.
The problem isn't the model. The problem is that every AI coding tool gives your AI a flat list of files and asks it to infer architecture from that. It can't. It guesses. And on non-trivial codebases, the guesses compound.
Specifically:\n- it doesn't know which functions depend on the file it's editing\n- it doesn't know your module ownership or layer rules\n- it doesn't know what changed since the last prompt\n- it has no way to validate its plan against your actual architecture before writing
Mikk fixes this by building a directed graph \u2014 the Mesh \u2014 of your entire codebase. Not a summary. Not embeddings. An actual graph: every function as a node, every import and call as a typed edge, forward and reverse adjacency maps for O(1) traversal in either direction.
`mikk init` runs in ~3 seconds on a 90-file project:\n- 2,847 functions parsed via TypeScript Compiler API (real AST, not regex)\n- 9,442 edges in the dependency graph\n- SHA-256 Merkle tree at function \u2192 file \u2192 module \u2192 root level\n- everything written to mikk.lock.json, ~60% smaller than raw source\n- claude.md + AGENTS.md generated with tiered architecture context
Then an MCP server exposes the graph. Your AI doesn't get a paste. It calls tools against the real graph mid-task.
Three things that actually matter:
1. blast radius before the edit\n`mikk_before_edit` walks the reverse dependency map from the file you're about to touch. Returns every upstream caller, every exported API at risk, every constraint that applies. Your AI sees the full blast radius before it writes a single line. `mikk_impact_analysis` goes deeper \u2014 transitive blast radius across the entire graph, estimated impact level, public API consumers.
2. context that's not RAG\nThe AI Context Builder seeds from your task description, BFS-traverses the call graph outward, scores each reachable function by proximity + keyword match + entry-point bonus, then greedily packs the highest-scoring functions into your token budget. Every result includes exact file path, start line, end line, and full source body. No hallucinated paths. No wrong modules.
3. architecture that enforces itself\nDefine module boundaries in mikk.json. Six constraint types: no-import, must-use, no-call, layer, naming, max-files. `mikk contract validate --strict` fails CI with exact file + rule citations. ADRs sit alongside constraints so your AI knows why a rule exists, not just that it does.
Local-first. No cloud. No telemetry. No API keys. TypeScript, JavaScript, Go.
https://github.com/ansh-dhanani/mikk\nhttps://mikk-web.vercel.app","title":"Mikk \u2013 your AI wrote the code but doesn't know what breaks","updated_at":"2026-03-15T14:42:03Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Ansh_Dhanani"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["vikk","ai"],"value":"The problem I keep hitting: AI assistants are making architectural decisions \u2014 what to import, which layer to touch, which module to put things in \u2014 without any understanding of the architecture they're working in.
It's not just wrong paths. It's more fundamental. The AI doesn't know your module ownership. It can't trace the blast radius of a change. It has no idea which constraints your team has defined or why. It gets files and hallucinates the rest.
I built Mikk to fix this \u2014 a codebase context engine that parses your TypeScript project and gives your AI the architecture it's missing. Dependency graph, module boundaries, blast radius analysis, architectural contracts, and a live MCP server so the AI can query your real architecture mid-task.
Curious how others are handling this. CLAUDE.md written by hand? RAG pipelines over your source? Context window tricks? What's actually working for non-trivial codebases?
(Mikk is open source: https://github.com/Ansh-dhanani/mikk)"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"How do you stop AI from making bad architectural decisions in your codebase?"}},"_tags":["story","author_Ansh_Dhanani","story_47261392","ask_hn"],"author":"Ansh_Dhanani","children":[47264158],"created_at":"2026-03-05T13:33:11Z","created_at_i":1772717591,"num_comments":3,"objectID":"47261392","points":2,"story_id":47261392,"story_text":"The problem I keep hitting: AI assistants are making architectural decisions \u2014 what to import, which layer to touch, which module to put things in \u2014 without any understanding of the architecture they're working in.
It's not just wrong paths. It's more fundamental. The AI doesn't know your module ownership. It can't trace the blast radius of a change. It has no idea which constraints your team has defined or why. It gets files and hallucinates the rest.
I built Mikk to fix this \u2014 a codebase context engine that parses your TypeScript project and gives your AI the architecture it's missing. Dependency graph, module boundaries, blast radius analysis, architectural contracts, and a live MCP server so the AI can query your real architecture mid-task.
Curious how others are handling this. CLAUDE.md written by hand? RAG pipelines over your source? Context window tricks? What's actually working for non-trivial codebases?
(Mikk is open source: https://github.com/Ansh-dhanani/mikk)","title":"How do you stop AI from making bad architectural decisions in your codebase?","updated_at":"2026-03-28T18:51:39Z"}],"hitsPerPage":50,"nbHits":11,"nbPages":1,"page":0,"params":"query=Vikk+AI&tags=story&hitsPerPage=50&advancedSyntax=true&analyticsTags=backend","processingTimeMS":7,"processingTimingsMS":{"_request":{"roundTrip":18},"afterFetch":{"format":{"highlighting":1,"total":1},"merge":{"total":1},"total":1},"fetch":{"query":2,"scanning":1,"total":4},"total":7},"query":"Vikk AI","serverTimeMS":9}