<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hacker News: leventov</title><link>https://news.ycombinator.com/user?id=leventov</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Thu, 19 Nov 2026 22:16:04 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=leventov" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[AOT compilation of Convex TypeScript functions into WASM]]></title><description><![CDATA[
<p>Article URL: <a href="https://convex-in-prod.github.io/posts/convex-wasm-compiler/">https://convex-in-prod.github.io/posts/convex-wasm-compiler/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50271059">https://news.ycombinator.com/item?id=50271059</a></p>
<p>Points: 2</p>
<p># Comments: 2</p>
]]></description><pubDate>Wed, 28 Oct 2026 09:38:57 +0000</pubDate><link>https://convex-in-prod.github.io/posts/convex-wasm-compiler/</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=50271059</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50271059</guid></item><item><title><![CDATA[Scaling self-hosted Convex: compilation of TypeScript Convex functions into WASM]]></title><description><![CDATA[
<p>Article URL: <a href="https://convex-in-prod.github.io/posts/convex-wasm-compiler/">https://convex-in-prod.github.io/posts/convex-wasm-compiler/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50249800">https://news.ycombinator.com/item?id=50249800</a></p>
<p>Points: 20</p>
<p># Comments: 0</p>
]]></description><pubDate>Mon, 26 Oct 2026 19:42:06 +0000</pubDate><link>https://convex-in-prod.github.io/posts/convex-wasm-compiler/</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=50249800</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50249800</guid></item><item><title><![CDATA[New comment by leventov in "Ask HN: What Are You Working On? (Nov 2025)"]]></title><description><![CDATA[
<p>I'm working on Pocketdata - a personal, private AI data plane.<p>The idea is to take boring components: PostgreSQL, Bifrost (LLM gateway), Open WebUI, LanceDB, Agentgateway (MCP and OpenAPI gateway) and deploy them in Fly.io. One Fly.io "org" per user. The closest equivalent is blaxel.ai, but it caters for AI SaaS startups, not individual customers.<p>The combination of the fact that Fly secrets are visible only from within the apps, distroless containers, and transparent data encryption for PostgreSQL assures that the service (Pocketdata) provider cannot access their data, only the infrastructure provider (Fly.io) theoretically can, but practically speaking, this gives an extremely high degree of privacy assurance.<p>The latest update on the project: <a href="https://engineeringideas.substack.com/p/tasklet-is-the-o1-moment-for-long" rel="nofollow">https://engineeringideas.substack.com/p/tasklet-is-the-o1-mo...</a></p>
]]></description><pubDate>Mon, 10 Nov 2025 07:08:01 +0000</pubDate><link>https://news.ycombinator.com/item?id=45873241</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=45873241</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=45873241</guid></item><item><title><![CDATA[New comment by leventov in "Let the little guys in: A context sharing runtime for the personalised web"]]></title><description><![CDATA[
<p>> It seems like a web version of claude code + skills + marketplace, but with an encrypted database, and permissions that let you feel safe using yolo mode. I’m going to try it.<p>I'm building infra for exactly this thing :) Here I posted about it today: <a href="https://engineeringideas.substack.com/p/tasklet-is-the-o1-moment-for-long" rel="nofollow">https://engineeringideas.substack.com/p/tasklet-is-the-o1-mo...</a><p>Architecture TLDR: In Fly.io, org-per user, manage keys in Fly secrets. A Postgres db with transparent data encryption (TDE), the master key is stored on user's computer in the keychain or in the password manager. Thus the nobody can read the data at rest. All containers are distroless so nobody can ssh onto them. Postgres is backed up via pgBackRest to Wasabi object storage with customer-provided encryption keys that are injected into the containers via Fly secrets.<p>Apart from the database for record-like things (chats, emails, tables), vectors for larger things (such as web pages) are stored in serverless LanceDB on Wasabi, too.<p>Also Bifrost (<a href="https://github.com/maximhq/bifrost" rel="nofollow">https://github.com/maximhq/bifrost</a>) as LLM gateway and Agentgateway (<a href="https://github.com/agentgateway/agentgateway" rel="nofollow">https://github.com/agentgateway/agentgateway</a>) as MCP and OpenAPI/REST API gateway.<p>Agents/apps themselves (Open WebUI, Zero Mail, etc.) are separate Fly apps, and have their separate schemas and users in Postgres. They also cannot go to public internet directly (prohibited via <a href="https://community.fly.io/t/new-feature-network-policies/19173">https://community.fly.io/t/new-feature-network-policies/1917...</a>), only to Postgres, Bifrost, and Agentgateway.<p>Postgres, pgBackRest, pgBouncer and another Go sidecar for pg (a la Pocketbase, but with Postgres backend), live in a single container managed by Horust. Bifrost and Agentgateway live in separate containers, but the same Fly machine. This machine might be 1.5gb, 4 vCPUs. LanceDB is on a separate machine because it needs burstable memory and may be infrequently used.<p>All machines (core, lance, and all individual app/agent machines) are suspendable, so they almost don't cost anything when not in use.</p>
]]></description><pubDate>Mon, 27 Oct 2025 21:05:03 +0000</pubDate><link>https://news.ycombinator.com/item?id=45726280</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=45726280</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=45726280</guid></item><item><title><![CDATA[Trace LLM workflows at your app's semantic level, not at the OpenAI API boundary]]></title><description><![CDATA[
<p>Article URL: <a href="https://engineeringideas.substack.com/p/trace-llm-workflows-at-your-apps">https://engineeringideas.substack.com/p/trace-llm-workflows-at-your-apps</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=44539301">https://news.ycombinator.com/item?id=44539301</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
]]></description><pubDate>Sat, 12 Jul 2025 04:25:58 +0000</pubDate><link>https://engineeringideas.substack.com/p/trace-llm-workflows-at-your-apps</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=44539301</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44539301</guid></item><item><title><![CDATA[Trace LLM workflows at your app's semantic level, not at the OpenAI API boundary]]></title><description><![CDATA[
<p>Article URL: <a href="https://engineeringideas.substack.com/p/trace-llm-workflows-at-your-apps">https://engineeringideas.substack.com/p/trace-llm-workflows-at-your-apps</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=44529295">https://news.ycombinator.com/item?id=44529295</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
]]></description><pubDate>Fri, 11 Jul 2025 07:29:37 +0000</pubDate><link>https://engineeringideas.substack.com/p/trace-llm-workflows-at-your-apps</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=44529295</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44529295</guid></item><item><title><![CDATA[New comment by leventov in "DuckDB 1.1.0 Released"]]></title><description><![CDATA[
<p>It's not super easy to port a library with domain-specific ML/DS algorithms built around Pandas, such as accepting Pandas dataframes as arguments to its functions and class constructors.</p>
]]></description><pubDate>Tue, 10 Sep 2024 06:57:11 +0000</pubDate><link>https://news.ycombinator.com/item?id=41497944</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=41497944</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=41497944</guid></item><item><title><![CDATA[New comment by leventov in "Predicting the Future of Distributed Systems"]]></title><description><![CDATA[
<p>BigQuery Storage Read API claims to support filters and simple projections pushed down to the storage: <a href="https://cloud.google.com/bigquery/docs/reference/storage" rel="nofollow">https://cloud.google.com/bigquery/docs/reference/storage</a>. See also this recent paper: <a href="https://research.google/pubs/biglake-bigquerys-evolution-toward-a-multi-cloud-lakehouse/" rel="nofollow">https://research.google/pubs/biglake-bigquerys-evolution-tow...</a><p>I've also recently proposed a Table Read protocol that should be a "non-vendor-controlled" equivalent of BigQuery Storage APIs: <a href="https://engineeringideas.substack.com/p/table-transfer-protocols-for-universal" rel="nofollow">https://engineeringideas.substack.com/p/table-transfer-proto...</a></p>
]]></description><pubDate>Sat, 31 Aug 2024 06:32:58 +0000</pubDate><link>https://news.ycombinator.com/item?id=41406939</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=41406939</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=41406939</guid></item><item><title><![CDATA[New comment by leventov in "Predicting the future of distributed systems"]]></title><description><![CDATA[
<p><a href="https://www.ray.io/" rel="nofollow">https://www.ray.io/</a></p>
]]></description><pubDate>Fri, 30 Aug 2024 08:16:36 +0000</pubDate><link>https://news.ycombinator.com/item?id=41398718</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=41398718</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=41398718</guid></item><item><title><![CDATA[New comment by leventov in "Predicting the future of distributed systems"]]></title><description><![CDATA[
<p>I don't buy this "object storage + Iceberg is all you need for OLAP" hype. If the application has sustained query load, it makes sense to provision servers rather than go pure serverless. If there are already provisioned servers, it makes sense to cache data on them (either in their memory or SSDs) to avoid round-trips to object storage. This is the architecture of the latest breed of OLAP databases such as Databend and Firebolt, as well as the latest iterations of Redshift's and Snowflake's architecture. Also, this is the approach of the newest breed of vector databases, such as Turbopuffer.<p>For OLAP use cases with real-time data ingestion requirements, object-storage-only approach also leads to write amplification. Therefore, I don't think that architectures like Apache Pinot, Apache Paimon, and Apache Druid are going anywhere.<p>Another problem with "open table formats" like Iceberg, Hudi, and Delta Lake is their slow innovation speed.<p>I've recently argued about this at greater length here: <a href="https://engineeringideas.substack.com/p/the-future-of-olap-table-storage" rel="nofollow">https://engineeringideas.substack.com/p/the-future-of-olap-t...</a></p>
]]></description><pubDate>Fri, 30 Aug 2024 06:27:26 +0000</pubDate><link>https://news.ycombinator.com/item?id=41398184</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=41398184</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=41398184</guid></item><item><title><![CDATA[Table transfer protocols for universal access to OLAP databases]]></title><description><![CDATA[
<p>Article URL: <a href="https://engineeringideas.substack.com/p/table-transfer-protocols-for-universal">https://engineeringideas.substack.com/p/table-transfer-protocols-for-universal</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=41360531">https://news.ycombinator.com/item?id=41360531</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
]]></description><pubDate>Mon, 26 Aug 2024 19:02:17 +0000</pubDate><link>https://engineeringideas.substack.com/p/table-transfer-protocols-for-universal</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=41360531</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=41360531</guid></item><item><title><![CDATA[New comment by leventov in "What happens in a mind that can't 'see' mental images"]]></title><description><![CDATA[
<p>Every once in a while, before falling asleep I enter a "mellow" state in which I can conjure mental images dramatically more vividly than normally. I would say that normally it varies between 10-50% of vividness (also depends on the state of mind, the day, whether I'm under the influence of different chemicals), and in this state of mind it's 90%+ vivid.<p>And I don't confuse this state of mind with lucid dreaming: I had spontaneous lucid dreams a few times. They were not very vivid, just as my normal dreams (I dream in images, and they are maybe around 50% vivid).<p>So, there is definitely a way to experience variation of vividness of mental imagery, by a single observer.</p>
]]></description><pubDate>Fri, 02 Aug 2024 22:58:59 +0000</pubDate><link>https://news.ycombinator.com/item?id=41143378</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=41143378</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=41143378</guid></item><item><title><![CDATA[The future of OLAP table storage is not Iceberg]]></title><description><![CDATA[
<p>Article URL: <a href="https://engineeringideas.substack.com/p/the-future-of-olap-table-storage">https://engineeringideas.substack.com/p/the-future-of-olap-table-storage</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=41112837">https://news.ycombinator.com/item?id=41112837</a></p>
<p>Points: 2</p>
<p># Comments: 0</p>
]]></description><pubDate>Tue, 30 Jul 2024 18:53:15 +0000</pubDate><link>https://engineeringideas.substack.com/p/the-future-of-olap-table-storage</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=41112837</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=41112837</guid></item><item><title><![CDATA[New comment by leventov in "AutoBNN: Probabilistic Time Series Forecasting"]]></title><description><![CDATA[
<p>Answering my own question about the other frameworks (one also from Google, BTW): <a href="https://exa.ai/search?c=news&q=time%20series%20prediction%20framework&filters=%7B%22domainFilterType%22%3A%22include%22%2C%22timeFilterOption%22%3A%22past_year%22%2C%22activeTabFilter%22%3A%22all%22%7D&autopromptString=">https://exa.ai/search?c=news&q=time%20series%20prediction%20...</a></p>
]]></description><pubDate>Sun, 31 Mar 2024 00:32:38 +0000</pubDate><link>https://news.ycombinator.com/item?id=39880143</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=39880143</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=39880143</guid></item><item><title><![CDATA[New comment by leventov in "AutoBNN: Probabilistic Time Series Forecasting"]]></title><description><![CDATA[
<p>What are the other four frameworks?<p>> For one, who wants to do stuff in tensorflow anymore let alone tensorflow-probability.<p>AutoBNN is a JAX library and has nothing to do technically with TF Probability. It was developed by the TF Probability <i>team</i>.<p>> DL community prefers pytorch and stats community prefers Stan.<p>It looks like the JAX ecosystem for stats is growing: NumPyro is based on JAX, PyMC has a JAX backend, <a href="https://github.com/blackjax-devs/blackjax">https://github.com/blackjax-devs/blackjax</a> has effective samplers, there is <a href="https://github.com/jax-ml/bayeux">https://github.com/jax-ml/bayeux</a>, and now AutoBNN.<p>> This one seems theoretically more interesting than some others but practically less useful.<p>Are there other factors why you think AutoBNN is not practically useful, apart from being based on the wrong foundation (which was a mistaken belief of yours)?</p>
]]></description><pubDate>Sun, 31 Mar 2024 00:25:54 +0000</pubDate><link>https://news.ycombinator.com/item?id=39880088</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=39880088</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=39880088</guid></item><item><title><![CDATA[New comment by leventov in "We reduced the cost of building Mastodon at Twitter-scale by 100x"]]></title><description><![CDATA[
<p>The big question is whether we really want more centralised cloud applications of this sort in the AI future, with its potentially exacerbated risks of data breaches, identity theft, spam, etc.<p>Having a "Rama" for local-first, truly distributed (Solid Project style), self-sovereign-identity based apps would be differentially better probably.</p>
]]></description><pubDate>Tue, 22 Aug 2023 11:04:00 +0000</pubDate><link>https://news.ycombinator.com/item?id=37221019</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=37221019</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=37221019</guid></item><item><title><![CDATA[New comment by leventov in "One AI Tutor Per Child: Personalized learning is finally here"]]></title><description><![CDATA[
<p>Bad analogy. Get in good shape to feel better, have more energy, be more attractive to other people, etc.<p>There is no point of getting good at arithmetic today, indeed.<p>I agree people need to continue study to understand the world (if they are to survive at all), but the current school curriculum is a joke.</p>
]]></description><pubDate>Fri, 17 Mar 2023 20:29:14 +0000</pubDate><link>https://news.ycombinator.com/item?id=35202469</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=35202469</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=35202469</guid></item><item><title><![CDATA[New comment by leventov in "GPT-4"]]></title><description><![CDATA[
<p>Ok, then what's the point of defining 'intelligence' out of existence, if neither LMs nor humans are 'intelligent'?</p>
]]></description><pubDate>Wed, 15 Mar 2023 10:47:27 +0000</pubDate><link>https://news.ycombinator.com/item?id=35166232</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=35166232</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=35166232</guid></item><item><title><![CDATA[New comment by leventov in "GPT-4"]]></title><description><![CDATA[
<p>It might be that humanity becoming non-chauvinistic about their value is the necessary thing for survival.</p>
]]></description><pubDate>Wed, 15 Mar 2023 10:41:31 +0000</pubDate><link>https://news.ycombinator.com/item?id=35166193</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=35166193</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=35166193</guid></item><item><title><![CDATA[New comment by leventov in "GPT-4"]]></title><description><![CDATA[
<p>You can't do it in your basement <i>as of 2023</i>. Very important qualification. It's entirely plausible that continuous evolution of ML architectures will lead to general AI which anyone can start on their phone and computer and learn online from there.</p>
]]></description><pubDate>Wed, 15 Mar 2023 10:38:14 +0000</pubDate><link>https://news.ycombinator.com/item?id=35166174</link><dc:creator>leventov</dc:creator><comments>https://news.ycombinator.com/item?id=35166174</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=35166174</guid></item></channel></rss>