Register allocation in the Go compiler (2024)
11 by elvis70 | 1 comments on Hacker News.
Tuesday, September 30, 2025
Monday, September 29, 2025
Sunday, September 28, 2025
Saturday, September 27, 2025
Friday, September 26, 2025
Thursday, September 25, 2025
Wednesday, September 24, 2025
Tuesday, September 23, 2025
New top story on Hacker News: Show HN: FlyCode – Recover Stripe payments by automatically using backup cards
Show HN: FlyCode – Recover Stripe payments by automatically using backup cards
11 by JakeVacovec | 26 comments on Hacker News.
We built FlyCode after seeing subscription businesses lose ~35% of recurring revenue each year to failed payments — even when customers had other valid cards on file. *The problem:* When a customer's primary card fails, Stripe retries a few times then cancels the subscription. If that customer has a backup card, it isn’t tried. At least 20% of active customers have more than one card on file, which means a lot of preventable churn. *Our solution:* FlyCode automatically identifies if a customer has other valid cards on file and retries them when a subscription payment fails. You can configure when these retries happen during the dunning period (beginning, middle, end) and define validity rules (e.g. “card was used in last 180 days”). It’s a Stripe app — no code changes needed. We've seen 18%-20% higher recovery rates from our core retry engine, plus another 5–10% from using backup cards. Importantly, there was no increase in refunds or chargebacks — in fact, rates were lower than merchant averages. Big companies like Microsoft and Amazon already do this internally; we wanted to make the same capability accessible to smaller SaaS teams. *Under the hood:* FlyCode monitors for failed invoices, checks for available backup methods via Stripe’s PaymentMethod API, and systematically retries in a way that avoids service disruption or manual workflows. We’re Jake, Etai, and Tzachi — we previously built payment recovery systems at startups and enterprises, which is how we discovered this gap. You can try it here: [ https://ift.tt/HQWejr3 ] We’d love feedback from anyone dealing with subscription payment failures. What’s been your experience with involuntary churn? Have you considered leveraging backup payment methods?
11 by JakeVacovec | 26 comments on Hacker News.
We built FlyCode after seeing subscription businesses lose ~35% of recurring revenue each year to failed payments — even when customers had other valid cards on file. *The problem:* When a customer's primary card fails, Stripe retries a few times then cancels the subscription. If that customer has a backup card, it isn’t tried. At least 20% of active customers have more than one card on file, which means a lot of preventable churn. *Our solution:* FlyCode automatically identifies if a customer has other valid cards on file and retries them when a subscription payment fails. You can configure when these retries happen during the dunning period (beginning, middle, end) and define validity rules (e.g. “card was used in last 180 days”). It’s a Stripe app — no code changes needed. We've seen 18%-20% higher recovery rates from our core retry engine, plus another 5–10% from using backup cards. Importantly, there was no increase in refunds or chargebacks — in fact, rates were lower than merchant averages. Big companies like Microsoft and Amazon already do this internally; we wanted to make the same capability accessible to smaller SaaS teams. *Under the hood:* FlyCode monitors for failed invoices, checks for available backup methods via Stripe’s PaymentMethod API, and systematically retries in a way that avoids service disruption or manual workflows. We’re Jake, Etai, and Tzachi — we previously built payment recovery systems at startups and enterprises, which is how we discovered this gap. You can try it here: [ https://ift.tt/HQWejr3 ] We’d love feedback from anyone dealing with subscription payment failures. What’s been your experience with involuntary churn? Have you considered leveraging backup payment methods?
Monday, September 22, 2025
New top story on Hacker News: Qwen3-Omni: Native Omni AI Model for Text, Image & Video
Qwen3-Omni: Native Omni AI Model for Text, Image & Video
19 by meetpateltech | 1 comments on Hacker News.
19 by meetpateltech | 1 comments on Hacker News.
Sunday, September 21, 2025
Saturday, September 20, 2025
Friday, September 19, 2025
Thursday, September 18, 2025
Wednesday, September 17, 2025
Tuesday, September 16, 2025
Monday, September 15, 2025
Sunday, September 14, 2025
Saturday, September 13, 2025
Friday, September 12, 2025
Thursday, September 11, 2025
Wednesday, September 10, 2025
New top story on Hacker News: Show HN: Haystack – Review pull requests like you wrote them yourself
Show HN: Haystack – Review pull requests like you wrote them yourself
21 by akshaysg | 7 comments on Hacker News.
Hi HN! We’re Akshay and Jake. We put together a tool called Haystack to make pull requests straightforward to read. What Haystack does: -- Builds a clear narrative. Changes in Haystack aren’t just arranged as unordered diffs. Instead, they unfold in a logical order, each paired with an explanation in plain, precise language -- Focuses attention where it counts. Routine plumbing and refactors are put into skimmable sections so you can spend your time on design and correctness -- Provides full cross-file context. Every new or changed function/variable is traced across the codebase, showing how it’s used beyond the immediate diff Here’s a quick demo: https://youtu.be/w5Lq5wBUS-I If you’d like to give it a spin, head over to haystackeditor.com/review! We set up some demo PRs that you should be able to understand and review even if you’ve never seen the repos before! We used to work at big companies, where reviewing non-trivial pull requests felt like reading a book with its pages out of order. We would jump and scroll between files, trying to piece together the author’s intent before we could even start reviewing. And, as authors, we would spend time to restructure our own commits just to make them readable. AI has made this even trickier. Today it’s not uncommon for a pull request to contain code the author doesn’t fully understand themselves! So, we built Haystack to help reviewers spend less time untangling code and more time giving meaningful feedback. We would love to hear about whether it gets the job done for you! How we got here: Haystack began as (yet another) VS Code fork where we experimented with visualizing code changes on a canvas. At first, it was a neat way to show how pieces of code worked together. But customers started laying out their entire codebase just to make sense of it. That’s when we realized the deeper problem: understanding a codebase is hard, and engineers need better ways to quickly understand unfamiliar code. As we kept building, another insight emerged: with AI woven into workflows, engineers don’t always need to master every corner of a codebase to ship features. But in code review, deep and continuous context still matters, especially to separate what’s important to review from plumbing and follow-on changes. So we pivoted. We took what we’d learned and worked closely with engineers to refine the idea. We started with simple code analysis (using language servers, tree-sitter, etc.) to show how changes relate. Then we added AI to explain and organize those changes and to trace how data moves through a pull request. Finally, we fused the two by empowering AI agents to use static analyses. Step by step, that became the Haystack we’re showing today. We’d love to hear your thoughts, feedback, or suggestions!
21 by akshaysg | 7 comments on Hacker News.
Hi HN! We’re Akshay and Jake. We put together a tool called Haystack to make pull requests straightforward to read. What Haystack does: -- Builds a clear narrative. Changes in Haystack aren’t just arranged as unordered diffs. Instead, they unfold in a logical order, each paired with an explanation in plain, precise language -- Focuses attention where it counts. Routine plumbing and refactors are put into skimmable sections so you can spend your time on design and correctness -- Provides full cross-file context. Every new or changed function/variable is traced across the codebase, showing how it’s used beyond the immediate diff Here’s a quick demo: https://youtu.be/w5Lq5wBUS-I If you’d like to give it a spin, head over to haystackeditor.com/review! We set up some demo PRs that you should be able to understand and review even if you’ve never seen the repos before! We used to work at big companies, where reviewing non-trivial pull requests felt like reading a book with its pages out of order. We would jump and scroll between files, trying to piece together the author’s intent before we could even start reviewing. And, as authors, we would spend time to restructure our own commits just to make them readable. AI has made this even trickier. Today it’s not uncommon for a pull request to contain code the author doesn’t fully understand themselves! So, we built Haystack to help reviewers spend less time untangling code and more time giving meaningful feedback. We would love to hear about whether it gets the job done for you! How we got here: Haystack began as (yet another) VS Code fork where we experimented with visualizing code changes on a canvas. At first, it was a neat way to show how pieces of code worked together. But customers started laying out their entire codebase just to make sense of it. That’s when we realized the deeper problem: understanding a codebase is hard, and engineers need better ways to quickly understand unfamiliar code. As we kept building, another insight emerged: with AI woven into workflows, engineers don’t always need to master every corner of a codebase to ship features. But in code review, deep and continuous context still matters, especially to separate what’s important to review from plumbing and follow-on changes. So we pivoted. We took what we’d learned and worked closely with engineers to refine the idea. We started with simple code analysis (using language servers, tree-sitter, etc.) to show how changes relate. Then we added AI to explain and organize those changes and to trace how data moves through a pull request. Finally, we fused the two by empowering AI agents to use static analyses. Step by step, that became the Haystack we’re showing today. We’d love to hear your thoughts, feedback, or suggestions!
New top story on Hacker News: Launch HN: Recall.ai (YC W20) – API for meeting recordings and transcripts
Launch HN: Recall.ai (YC W20) – API for meeting recordings and transcripts
18 by davidgu | 8 comments on Hacker News.
Hey HN, we're David and Amanda from Recall.ai ( https://www.recall.ai ). Today we’re launching our Desktop Recording SDK, a way to get meeting data without a bot in the meeting: https://ift.tt/Ha1oxML . It’s our biggest release in quite a while so we thought we’d finally do our Launch HN :) Here’s a demo that shows it producing a transcript from a meeting, followed by examples in code: https://www.youtube.com/watch?v=4croAGGiKTA . API docs are at https://docs.recall.ai/ . Back in W20, our first product was an API that lets you send a bot participant into a meeting. This gives developers access to audio/video streams and other data in the meeting. Today, this API powers most of the meeting recording products on the market. Recently, meeting recording through a desktop form factor instead of a bot has become popular. Many products like Notion and ChatGPT have added desktop recording functionality, and LLMs have made it easier to work with unstructured transcripts. But it’s actually hard to reliably record meetings at scale with a desktop app, and most developers who want to add recording functionality don’t want to build all this infrastructure. Doing a basic recording with just the microphone and system audio is fairly straightforward since you can just use the system APIs. But it gets a lot harder when you want to capture speaker names, produce a video recording, get real-time data, or run this in production at large scale: - Capturing speaker names involves using accessibility APIs to screen-scrape the video conference window to monitor who is speaking at what time. When video conferencing platforms change their UI, we must ship a change immediately, so this keeps working. - Producing a video recording that is clean, and doesn’t capture the video conferencing platform UI involves detecting the participant tiles, cropping them out, and compositing them together into a clean video recording. - Because the desktop recording code runs on end-user machines, we need to make it as efficient as possible. This means writing highly platform-optimized code, taking advantage of hardware encoders when available, and spending a lot of time doing profiling and performance testing. Meeting recording has zero margin for failure because if anything breaks, you lose the data forever. Reliability is especially important, which dramatically increases the amount of engineering effort required. Our Desktop Recording SDK takes care of all this and lets developers build meeting recording features into their desktop apps, so they can record both video conferences and in-person meetings without a bot. We built Recall.ai because we experienced this problem ourselves. At our first startup, we built a tool for product managers that included a meeting recording feature. 70% of our engineering time was taken up by just this feature! We ended up starting Recall.ai to solve this instead. Since then, over 2000 companies use us to power their recording features, e.g. Hubspot for sales call recording, Clickup for their AI note taker. Our users are engineering teams building commercial products for financial services, telehealth, incident management, sales, interviewing, and more. We also power internal tooling for large enterprises. Running this sort of infrastructure has led to unexpected technical challenges! For example, we had to debug a 1 in 36 million segfault in our audio encoder ( https://ift.tt/oxurgyG... ), we encountered a Postgres lock-up that only occurs when you have tens of thousands of concurrent writers ( https://ift.tt/2T3nEPH ), and we saved over $1M a year on AWS by optimizing the way we shuffle data around between our processes ( https://ift.tt/T0zmP6R ). You can try it here: https://www.recall.ai . It's self-serve with $5 of free credits. Pricing starts at $0.70 for every hour of recording, prorated to the second. We offer volume discounts with scale. All data recorded through Recall.ai is the property of our customers, we support 0-day retention, and we don’t train models on customer data. We would love your feedback!
18 by davidgu | 8 comments on Hacker News.
Hey HN, we're David and Amanda from Recall.ai ( https://www.recall.ai ). Today we’re launching our Desktop Recording SDK, a way to get meeting data without a bot in the meeting: https://ift.tt/Ha1oxML . It’s our biggest release in quite a while so we thought we’d finally do our Launch HN :) Here’s a demo that shows it producing a transcript from a meeting, followed by examples in code: https://www.youtube.com/watch?v=4croAGGiKTA . API docs are at https://docs.recall.ai/ . Back in W20, our first product was an API that lets you send a bot participant into a meeting. This gives developers access to audio/video streams and other data in the meeting. Today, this API powers most of the meeting recording products on the market. Recently, meeting recording through a desktop form factor instead of a bot has become popular. Many products like Notion and ChatGPT have added desktop recording functionality, and LLMs have made it easier to work with unstructured transcripts. But it’s actually hard to reliably record meetings at scale with a desktop app, and most developers who want to add recording functionality don’t want to build all this infrastructure. Doing a basic recording with just the microphone and system audio is fairly straightforward since you can just use the system APIs. But it gets a lot harder when you want to capture speaker names, produce a video recording, get real-time data, or run this in production at large scale: - Capturing speaker names involves using accessibility APIs to screen-scrape the video conference window to monitor who is speaking at what time. When video conferencing platforms change their UI, we must ship a change immediately, so this keeps working. - Producing a video recording that is clean, and doesn’t capture the video conferencing platform UI involves detecting the participant tiles, cropping them out, and compositing them together into a clean video recording. - Because the desktop recording code runs on end-user machines, we need to make it as efficient as possible. This means writing highly platform-optimized code, taking advantage of hardware encoders when available, and spending a lot of time doing profiling and performance testing. Meeting recording has zero margin for failure because if anything breaks, you lose the data forever. Reliability is especially important, which dramatically increases the amount of engineering effort required. Our Desktop Recording SDK takes care of all this and lets developers build meeting recording features into their desktop apps, so they can record both video conferences and in-person meetings without a bot. We built Recall.ai because we experienced this problem ourselves. At our first startup, we built a tool for product managers that included a meeting recording feature. 70% of our engineering time was taken up by just this feature! We ended up starting Recall.ai to solve this instead. Since then, over 2000 companies use us to power their recording features, e.g. Hubspot for sales call recording, Clickup for their AI note taker. Our users are engineering teams building commercial products for financial services, telehealth, incident management, sales, interviewing, and more. We also power internal tooling for large enterprises. Running this sort of infrastructure has led to unexpected technical challenges! For example, we had to debug a 1 in 36 million segfault in our audio encoder ( https://ift.tt/oxurgyG... ), we encountered a Postgres lock-up that only occurs when you have tens of thousands of concurrent writers ( https://ift.tt/2T3nEPH ), and we saved over $1M a year on AWS by optimizing the way we shuffle data around between our processes ( https://ift.tt/T0zmP6R ). You can try it here: https://www.recall.ai . It's self-serve with $5 of free credits. Pricing starts at $0.70 for every hour of recording, prorated to the second. We offer volume discounts with scale. All data recorded through Recall.ai is the property of our customers, we support 0-day retention, and we don’t train models on customer data. We would love your feedback!
Tuesday, September 9, 2025
Monday, September 8, 2025
Sunday, September 7, 2025
Saturday, September 6, 2025
Friday, September 5, 2025
New top story on Hacker News: Show HN: Open-sourcing our text-to-CAD app
Show HN: Open-sourcing our text-to-CAD app
23 by zachdive | 4 comments on Hacker News.
Hey HN! I'm Zach from Adam ( https://adam.new/ ). We’re building an AI co-pilot for mechanical CAD software. As part of our broader research, we built a browser-based Text-to-CAD app ( https://ift.tt/RF8YrJx ) and are now open sourcing it. This is a React SPA with a Supabase backend. What it does: * Generates parametric 3D models from natural language descriptions, with support for both text prompts and image references * Outputs OpenSCAD code with automatically extracted parameters that surface as interactive sliders for instant dimension tweaking * Exports as .STL or .SCAD Under the hood: * Separate agents for conversation and code generation; simple parameter tweaks bypass AI entirely using deterministic regex-based updates * Runs fully in-browser by compiling OpenSCAD to WebAssembly and integrating Three.js with React Three Fiber for 3D rendering * Supports BOSL, BOSL2, MCAD libraries and custom font support (Geist) for text in models We’ve seen many developers trying to replicate this kind of functionality, so we’re releasing this to give the community a solid foundation to build on. Future improvements: * Expand geometry support - Move beyond CSG primitives to support curved surfaces, fillets, lofts, and constraint-driven modeling through CadQuery/Build123D * Better spatial context - UI for face/edge selection and viewport image integration to give LLMs spatial understanding * Enhanced capabilities - RAG on documentation and integration with more OpenSCAD libraries for features like proper threading You can clone the repo and run it locally! Contributions are welcome, and we’ll keep merging PRs as they come in.
23 by zachdive | 4 comments on Hacker News.
Hey HN! I'm Zach from Adam ( https://adam.new/ ). We’re building an AI co-pilot for mechanical CAD software. As part of our broader research, we built a browser-based Text-to-CAD app ( https://ift.tt/RF8YrJx ) and are now open sourcing it. This is a React SPA with a Supabase backend. What it does: * Generates parametric 3D models from natural language descriptions, with support for both text prompts and image references * Outputs OpenSCAD code with automatically extracted parameters that surface as interactive sliders for instant dimension tweaking * Exports as .STL or .SCAD Under the hood: * Separate agents for conversation and code generation; simple parameter tweaks bypass AI entirely using deterministic regex-based updates * Runs fully in-browser by compiling OpenSCAD to WebAssembly and integrating Three.js with React Three Fiber for 3D rendering * Supports BOSL, BOSL2, MCAD libraries and custom font support (Geist) for text in models We’ve seen many developers trying to replicate this kind of functionality, so we’re releasing this to give the community a solid foundation to build on. Future improvements: * Expand geometry support - Move beyond CSG primitives to support curved surfaces, fillets, lofts, and constraint-driven modeling through CadQuery/Build123D * Better spatial context - UI for face/edge selection and viewport image integration to give LLMs spatial understanding * Enhanced capabilities - RAG on documentation and integration with more OpenSCAD libraries for features like proper threading You can clone the repo and run it locally! Contributions are welcome, and we’ll keep merging PRs as they come in.
Thursday, September 4, 2025
New top story on Hacker News: A high schooler writes about AI tools in the classroom
A high schooler writes about AI tools in the classroom
91 by dougb5 | 79 comments on Hacker News.
https://ift.tt/DyLJXeW
91 by dougb5 | 79 comments on Hacker News.
https://ift.tt/DyLJXeW
Wednesday, September 3, 2025
New top story on Hacker News: Show HN: Entropy-Guided Loop – How to make small models reason
Show HN: Entropy-Guided Loop – How to make small models reason
13 by andrewmonostate | 0 comments on Hacker News.
TLDR: A small, vendor-agnostic inference loop that turns token logprobs/perplexity/entropy into an extra pass and reasoning for LLMs. - Captures logprobs/top-k during generation, computes perplexity and token-level entropy. - Triggers at most one refine when simple thresholds fire; passes a compact “uncertainty report” (uncertain tokens + top-k alts + local context) back to the model. - In our tests on technical Q&A / math / code, a small model recovered much of “reasoning” quality at ~⅓ the cost while refining ~⅓ of outputs. I kept seeing “reasoning” models behave like expensive black boxes. Meanwhile, standard inference already computes useful signals both before softmax normalization and after it(logprobs), which we usually throw away. This loop tries the simplest thing that you could think of: use those signals to decide when (and where) to think again. GitHub (notebook + minimal code): https://ift.tt/2h0qbFR Paper (short & engineer made): https://ift.tt/KLFgRd3 Blog (more context): https://ift.tt/70motVd Requirements: Python, API that exposes logprobs (tested with OpenAI non reasoning 4.1). OPENAI_API_KEY and WEAVE for observability. Run the notebook; it prints metrics and shows which tokens triggered refinement. - Python, simple loop (no retraining). - Uses Responses API logprobs/top-k; metrics: perplexity, max token entropy, low-confidence counts. - Weave for lightweight logging/observability (optional). - Passing alternatives (not just “this looks uncertain”) prevents over-correction. - A simple OR rule (ppl / max-entropy / low-confidence count) catches complementary failure modes. - Numbers drift across vendors; keeping the method vendor-agnostic is better than chasing fragile pairings. - Needs APIs that expose logprobs/top-k. - Results are indicative—not a leaderboard; focus is on within-model gains (single-pass vs +loop). - Thresholds might need light tuning per domain. - One pass only; not a chain-of-thought replacement. - Run it on your models and ideas (e.g., 4o-mini, v3, Llama variants with logprobs) and share logs in a PR for our README in GitHub if you'd like, PRs welcome - I’ll credit and link. Overall let me know if you find making small models reason like this useful!
13 by andrewmonostate | 0 comments on Hacker News.
TLDR: A small, vendor-agnostic inference loop that turns token logprobs/perplexity/entropy into an extra pass and reasoning for LLMs. - Captures logprobs/top-k during generation, computes perplexity and token-level entropy. - Triggers at most one refine when simple thresholds fire; passes a compact “uncertainty report” (uncertain tokens + top-k alts + local context) back to the model. - In our tests on technical Q&A / math / code, a small model recovered much of “reasoning” quality at ~⅓ the cost while refining ~⅓ of outputs. I kept seeing “reasoning” models behave like expensive black boxes. Meanwhile, standard inference already computes useful signals both before softmax normalization and after it(logprobs), which we usually throw away. This loop tries the simplest thing that you could think of: use those signals to decide when (and where) to think again. GitHub (notebook + minimal code): https://ift.tt/2h0qbFR Paper (short & engineer made): https://ift.tt/KLFgRd3 Blog (more context): https://ift.tt/70motVd Requirements: Python, API that exposes logprobs (tested with OpenAI non reasoning 4.1). OPENAI_API_KEY and WEAVE for observability. Run the notebook; it prints metrics and shows which tokens triggered refinement. - Python, simple loop (no retraining). - Uses Responses API logprobs/top-k; metrics: perplexity, max token entropy, low-confidence counts. - Weave for lightweight logging/observability (optional). - Passing alternatives (not just “this looks uncertain”) prevents over-correction. - A simple OR rule (ppl / max-entropy / low-confidence count) catches complementary failure modes. - Numbers drift across vendors; keeping the method vendor-agnostic is better than chasing fragile pairings. - Needs APIs that expose logprobs/top-k. - Results are indicative—not a leaderboard; focus is on within-model gains (single-pass vs +loop). - Thresholds might need light tuning per domain. - One pass only; not a chain-of-thought replacement. - Run it on your models and ideas (e.g., 4o-mini, v3, Llama variants with logprobs) and share logs in a PR for our README in GitHub if you'd like, PRs welcome - I’ll credit and link. Overall let me know if you find making small models reason like this useful!
Tuesday, September 2, 2025
Monday, September 1, 2025
Subscribe to:
Comments (Atom)