www.dbaman.net
Show HN: Wick – censorship circumvention tech repurposed for AI web access
I helped build Lantern, a censorship circumvention tool that’s been downloaded over 150 million times globally, with most users in China, Russia, Iran, and the UAE. Before Lantern I was the lead engineer at LimeWire (back in those days our whole dev team was about 5 people!).Making your traffic indistinguishable from normal internet traffic has long been a foundational part of effective censorship circumvention, a technique sometimes referred to as “collateral freedom.” If censors can’t differen
Why Most Valuable AI Systems Are Still Tabular Models
The Hard Part of Predictive AI Isn’t the ModelI’ve spent most of my career building predictive systems on tabular data.The highest-value AI systems I’ve seen in production aren’t LLMs. They’re predictive models that operate on structured operational data: customers, orders, shipments, transactions, support events, etc.These systems quietly generate millions in value by replacing expensive third-party services, improving operational decisions, and turning predictions into products.Examples includ
Show HN: Gemma 4 based local RAG on 25 Years of news articles
A fully local Retrieval-Augmented Generation (RAG) implementation for querying 25 years of Swiss Teletext news (~500k articles in German language) - based on Deepmind's most recent Gemma model.<p>Why? I thought it's a cool type of dataset (short/high density news summaries) to test some local RAG approaches. Gemma 4 gives some impressive results, but could probably use some more tweaking on the system prompt.
ARCHE3-7B – Sparse Moe with SmartRouter and Foundation Curriculum Training
This is my first post on HN — a bit nervous, but excited to share what I've been building.I’ve been working on a 7B sparse Mixture-of-Experts prototype that can actually run on consumer hardware. For example, on a Colab T4 it uses around 5 GB RAM and 5 GB VRAM during training, and roughly 3.5–5 GB for inference.A couple of things I spent a lot of time on:Routing (SmartRouter)
I tried to tackle routing collapse in a practical way. Instead of letting all tokens dump into a few "favorite&
Show HN: PhAIL – Real-robot benchmark for AI models
I built this because I couldn't find honest numbers on how well VLA models [1] actually work on commercial tasks. I come from search ranking at Google where you measure everything, and in robotics nobody seemed to know.PhAIL runs four models (OpenPI/pi0.5, GR00T, ACT, SmolVLA) on bin-to-bin order picking – one of the most common warehouse operations. Same robot (Franka FR3), same objects, hundreds of blind runs. The operator doesn't know which model is running.Best model: 64 UPH.
Show HN: Pre-training, fine-tuning, and evals platform
Since yesterday, I've created a dataset out of a few docs I had (hackathon organization instructions) and then started fine tuning qwen 3.5. So it only took a few hours! I'm impressed by the ease of use - it's a chatbot and even if you don't have a lot of experience you can follow steps. Though don't want to undermine it, I'm sure could be a great fit for serious labs too.
Ask HN: Is there any interest in a native Qt/C++ Discord client?
I've been building a third-party cross-platform Discord client called kind (kind is not discord) in C++ and Qt 6. No Electron, no web wrapper. A proper native app for Windows, macOS, and Linux.The UI always loads instantly. On first load, things take as long to appear as Discord's network responses. On subsequent loads, everything comes from cache and is validated against REST in the background. It feels as fast as opening a plaintext file.It's further along than I expected when I
Show HN: Castra – Strip orchestration rights from your LLMs
I got tired of AI agents forgetting what they were doing the moment their context window filled. The current industry solution is to write massively bloated agent harnesses full of defensive spaghetti just to stop models from drifting.The problem is treating chat history as project state. A conversation is not a ledger.Castra is a compiled Go binary that strips orchestration rights from the LLM. State lives in an encrypted, local SQLite database (castra.db). The LLM is just a stateless executor
We scanned 73 open-source MCP servers. Here's what source code analysis found
We ran Sigil, our open-source security scanner, against 73 of the most-installed MCP servers on Smithery. Every finding was manually verified against the source code.Results: 66 passed (90%). 5 had configuration warnings. 2 failed with real security-relevant patterns in their tool handlers.The two failures are instructive:telegram-mcp scored F with 12 findings. Its send_photo, send_video, and send_document tools accept a file_path parameter with zero path validation and no directory allowlist. A
Show HN: Quillium, the non-linear writing app
Public beta lauunch!I built a SvelteKit + Tauri writing app that treats revision as a first-class concept. You can fork any sentence or passage into branches, switch between versions inline, and keep everything in one document instead of maintaining separate draft files.It's different from Git in the sense that you can try different combinations of these branches AND that these branches go infinitely nestedThe editor component was made with CodeMirror, and I tried really hard with implement
Show HN: SparrowDB, SQLite of graph DB, written in Rust, Cypher compatible
SparrowDB is an embedded Cypher graph database written in Rust. It links into your process the way SQLite does.I built it because I wanted a graph DB without running Docker or a cloud instance.SparrowDB kicks butt at point lookups, global COUNT(*), and top-10 by degree but lags on traversals (1-hop, 2-hop, and variable-length paths).It's an embedded engine with no parallel execution vs a warmed JVM server with a mature query planner.
The traversal workload is not the target. Agents, CLIs, s
Show HN: Matrix OS, like Lovable, but for personal apps
hey hn, i built matrix os, a personal ai operating system that generates custom software from natural language.you get your own cloud instance at matrix-os.com. you describe what you want ("build me an expense tracker with categories") and it appears on your desktop as a real app saved as a file.tech stack: node.js, typescript, claude agent sdk as the kernel, next.js frontend, hono gateway, sqlite/drizzle. everything is a file, apps, data, settings, ai memory. git-versioned.what m
Show HN: AptSelect – A local desktop app to test LLMs side-by-side
Hi HN,Whenever I needed an LLM to reliably output JSON or follow strict formatting rules, I kept having to write throwaway JavaScript scripts just to test the same prompt against OpenAI, Anthropic, and Google APIs to see who actually followed the instructions. It was a tedious workflow, so I built a local desktop UI to just do it for me.What it does:* Sends one prompt to OpenAI, Anthropic, Mistral, and Google simultaneously so you can compare the outputs in a single grid.* Automatically checkpoi
WebAssembly, the Safer Alternative to Integrating Native Code in Java
A monthly overview of things you need to know as an architect or aspiring architect. Unlock the full InfoQ experience by logging in! Stay updated with your favorite authors and topics, engage with ...
Stream live data from Amazon Keyspaces to S3 vector for real time AI applications
The data freshness challenge in artificial intelligence applications Large language models (LLMs) are trained over extended periods using expensive compute resources, resulting in knowledge that becomes stale over time. This creates a significant gap between current information and what AI systems can access. This limitation becomes critical when organizations need AI applications that understand and respond to real-time business events, current industry conditions, or recent data changes. You c
Turbocharge your applications with Amazon DocumentDB 8.0
Amazon DocumentDB (with MongoDB compatibility) announced the general availability of Amazon DocumentDB 8.0 that delivers breakthrough performance improvements that can transform your application experience. With up to 7x faster aggregation pipeline latency and 5x improved storage compression, you can build faster applications while significantly reducing costs. Amazon DocumentDB 8.0 brings in support for MongoDB 8.0 API driver compatibility while maintaining support for applications built using
Conversational Oracle EBS operations with CloudWatch MCP and Kiro CLI
Imagine asking your Oracle E-Business Suite (Oracle EBS) environment a question in plain English—”Is the concurrent manager healthy?”—and getting an immediate, comprehensive answer that pulls data from your monitoring infrastructure, application logs, and system status checks. This is possible by combining Kiro CLI, an AI-powered command line assistant, with Amazon CloudWatch through the Model Context Protocol (MCP), a standard that enables AI assistants to securely access external data sources
EDB Postgres AI for WarehousePG: Reclaiming control of the enterprise data warehouse
Proprietary warehouses delivered scale — but at the cost of control, predictable pricing, and real flexibility. Enterprises are doing the math.
Migrating to Amazon ElastiCache for Valkey: Best practices and a customer success story
Amazon ElastiCache for Valkey is a fully managed, high-performance, and cost-effective in-memory caching solution that serves as a drop-in alternative to Redis OSS. As an open source fork of Redis OSS 7.2.4, Valkey is maintained under the permissive BSD 3-clause license, addressing concerns arising from Redis OSS’s transition from BSD-3-Clause to a dual-license model (RSALv2 and SSPLv1) in March 2024, which prompted the creation of Valkey as an open-source fork. In this post, we provide a guide
Show HN: Postgres extension for BM25 relevance-ranked full-text search
Last summer we faced a conundrum at my company, Tiger Data, a Postgres cloud vendor whose main business is in timeseries data. We were trying to grow our business towards emerging AI-centric workloads and wanted to provide a state-of-the-art hybrid search stack in Postgres. We'd already built pgvectorscale in house with the goal of scaling semantic search beyond pgvector's main memory limitations. We just needed a scalable ranked keyword search solution too.The problem: core Postgre