AI Developmentr/LocalLLaMA

I classified 3.5M US patents with Nemotron 9B on a single RTX 5090 — then built a free search engine on top

Read original
ai-coding-toolslocal-llm-workflowsdomain-specific-aihybrid-search-architecture

“Patent attorneys need exact phrase matching. 'solid-state battery electrolyte' should match those exact words, not semantically similar documents about 'energy storage.' FTS5 gives sub-second queries on 3.5M records with zero external dependencies.”

Key takeaways

  • Local LLM (Nemotron 9B on RTX 5090) classified 3.5M patents in 48 hours, demonstrating consumer hardware can handle enterprise-scale classification tasks
  • Contrarian architecture choice: FTS5 full-text search outperforms vector embeddings for domain-specific use cases requiring exact phrase matching and deterministic results
  • Complete technical stack disclosed: SQLite FTS5 + local LLM query expansion + BM25 ranking with custom weights + FastAPI, hosted on Chromebook via Cloudflare Tunnel - proving production-grade search doesn't require cloud infrastructure
  • Patent lawyer with 1 month coding experience built production search engine, signaling democratization of AI tooling for domain experts without traditional engineering backgrounds
  • Hybrid approach: Uses LLM for natural language query expansion into boolean queries, then traditional search for retrieval - combining strengths of both paradigms

Why this matters for operators: Legal tech, enterprise search architecture, local LLM deployment for specialized domains, hybrid search strategies

I cover AI×GTM intelligence like this every Wednesday.

Get STEEPWORKS Weekly

More picks

GTM OpsDemand Gen ReportVictor's pick

Trust is the New Currency in B2B Buying: SurveyMonkey, Reddit

These are high % stats showing what we implicitly already know

  • Peer validation (73% trust) now dramatically outweighs traditional vendor marketing (55% trust vendor sites, 39% trust AI chatbots, 36% trust social media) in early-stage B2B buying
  • 83% of B2B buyers complete self-directed research before sales engagement, with high-stakes categories (software, professional services, HR) taking several weeks to months in extended evaluation
  • Search engines serve as navigation layer, not destination—buyers use search to identify options then validate through peer communities like Reddit (121M daily users, 19% YoY growth), creating imperative for authentic community presence
community-led-growthback-to-basics-gtmhuman-first-sales
Read original Full analysis →
AI DevelopmentGTM AI Podcast & NewsletterVictor's pick

Claude Channels

The move from user initiated to automated workflows is one of the main transitions with current agentic capabilities IMO

  • Claude Channels (launched March 20, 2026) enables event-driven AI automation via MCP protocol, shifting from pull-based (user-initiated) to push-based (event-triggered) workflows
  • Practical use case: CI/CD failures can trigger autonomous investigation, fix deployment, and resolution without human intervention - reducing 12-hour incident windows to near-zero
  • Technical implementation uses MCP servers connecting Claude Code to messaging platforms (Telegram/Discord at launch), with Bun runtime for 4x faster cold-start performance vs Node
ai-coding-toolsautomation-stackssignal-infrastructure
Read original Full analysis →
AI×GTMThe InformationVictor's pick

AWS Accelerates Internal AI Agents Following Staff Cuts

If you think white collar job displacement is a joke, or a distant future concern, this is just one more sign it is most definitely NOT. It's here.

  • AWS is deploying AI agents to handle technical sales support functions previously performed by thousands of specialists
  • The AI automation directly correlates with recent layoffs of hundreds in sales, business development, and technical specialist roles
  • Major cloud provider is using its own AI capabilities to reduce headcount in customer-facing technical roles, signaling broader industry trend
ai-sdr-adoptionautomation-stacksback-to-basics-gtm
Read original Full analysis →

This analysis was produced using the STEEPWORKS system — the same agents, skills, and knowledge architecture available in the GrowthOS package.