We built a native inference engine that runs quantized LLMs directly from SSD using memory-mapped I/O. The model never fully loads into RAM — the OS pages weights on demand as each layer executes.*What it does:* - Mixtral 8x22B (80GB, 141B params) runs on a machine with 48GB RAM - Model loads in 0.3 seconds (vs 190s with llama.cpp) - Produces correct output: "What is 2+2?" → "The sum of 2 and 2...
RLM-MCP – Analyze massive files in Claude Code using MIT's Recursive Language Models.I built an MCP server that lets Claude Code analyze files that exceed its context window by implementing the Recursive Language Models approach from MIT (arXiv:2512.24601).The problem: Claude Code can't fit a 10GB log file in its context. Traditional grep/read returns thousands of tokens of raw matches.The...
Hey HN, I’m the solo builder behind LogiCart.I recently refactored my frontend to use a Generative UI pattern (inspired by Google's new A2UI framework) because I realized a static chat interface fails for complex shopping intents.The Problem: A user buying a single item needs a completely different UX than a user planning a complex project. A standard "list of cards" doesn't work for both.The...
The Problem: Keyword search fails for complex, multi-item projects. If I want to "build a backyard rink" or "draft-proof my house," I usually end up with 15 tabs open, manually piecing together a shopping list from forums and reviews.
The Solution: I built an intent-based search engine. instead of matching keywords, it converts user prompts into vector embeddings to map broad goals (intents) to specific product clusters.
The Stack:
Backend: node.js/TypeScript
Search: pgvector for semantic similarity (PostgreSQL)
LLM: Orchestrates the "logic" of the project plan and validates compatibility.
The "Microsoft" Context: I actually launched this 24 hours before Microsoft Copilot released their "Shopping Agent." While I can't compete with them on generic "buy a laptop" queries, I’m pivoting to go deep on complex DIY/Hardware projects where context matters more than reach.
Current Status: Currently indexing Amazon (US/CA) with every search. My internal prototype worked perfectly with Canadian Tire (local hardware inventory is much better for DIY than Amazon), and I'm working on deep-linking that next to build better carts.
I’d love feedback on the result relevance. Roast my MVP.