> Markdown version of [/videos/100646-the-feature-that-didn-t-need-to-be-smart?t=42](https://www.wearedevelopers.com/videos/100646-the-feature-that-didn-t-need-to-be-smart?t=42). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md). --- # The Feature That Didn’t Need to Be Smart How do you build a genuinely useful LLM feature without relying on cloud APIs or sacrificing user privacy? See how a skeptical developer tamed local open-weight models using strict JSON validation and zero-temperature retries. - **Speakers:** Carmen Huidobro - **Event:** HalfStack Vienna 2026 - **Published:** November 19, 2026 - **Duration:** 36:12 - **URL:** https://www.wearedevelopers.com/videos/100646-the-feature-that-didn-t-need-to-be-smart - **Thumbnail URL:** https://customer-goifxnzhrq3d0d54.cloudflarestream.com/09d955e1a13fda88091ac928463ac818/thumbnails/thumbnail.jpg?time=21300ms ## Summary Facing industry burnout and skepticism toward generative AI hype, veteran developer Carmen H. returned to commercial indie software to work on Renamer, an Electron-based batch file renaming app. Tasked with integrating AI into the product, she approached Large Language Models (LLMs) objectively as a narrow utility rather than a magical cure-all. By reflecting on historical customer support queries, she realized the ideal application was not letting AI directly manipulate user files, but rather helping users author complex renaming rule chains—called "renamerlets"—from natural language descriptions. To align with strict ethical standards, the resulting "AI Assist" feature was designed to be completely private, local, opt-in, and free of cloud subscriptions. Instead of relying on proprietary cloud APIs, the architecture allows users to download and run an open-weight Mistral model locally via the Llama CLI. The workflow normalizes multilingual user requests into English, infers intent alongside existing filenames, and enforces a strict JSON schema to output valid action arrays. Choosing JSON over YAML was a critical insight, as YAML parsers notoriously misinterpret unquoted strings like "on" or "off" as booleans. Because LLMs frequently hallucinate or act overly conversational, building a resilient validation and graceful failure pipeline was essential. The application validates the generated JSON against allowed application actions and, if malformed, silently prompts the model to retry with the temperature set to zero to eliminate inventiveness. Furthermore, investing in a 30-hour evaluation corpus of unit tests ensured the prompt combinations reliably translated messy human intent into deterministic code. Ultimately, while local models introduce hardware trade-offs like a hefty 1.8 GB binary download and high CPU usage on older Intel chips, treating AI strictly as a translation layer proves that "it is possible to add AI in such a way that can be an interface to something you already trust." **Keywords:** local llm deployment, mistral ai model, llama cli integration, ethical ai development, json schema validation, natural language to json, electron app development, batch file renaming, llm evaluation corpus, graceful ai failure handling, prompt normalization workflow, yaml boolean parsing issues, opt-in ai features, local ai privacy, llm temperature tuning ## Chapters 1. **Early career in Mac OS X and developer relations** (00:42) — Navigating a twenty-year software engineering career from early Objective-C and Subversion days to Ruby on Rails, Rust, and developer education. 1. **Navigating the uncanny valley of generative AI** (03:58) — Comparing modern generative AI hype and industry burnout to historical computing milestones like the 1961 IBM singing computer. 1. **Scoping an AI feature for a batch file renamer** (09:43) — Adding an LLM integration to an Electron batch file renaming app by focusing on discrete user assistance rather than reinventing the core workflow. 1. **Establishing ethical boundaries and JSON schemas for AI** (12:38) — Enforcing ethical software standards by using local open-weight models like Mistral, making AI opt-in, and translating natural language into strict JSON instead of YAML. 1. **Designing the multi-step prompt and Llama CLI workflow** (17:46) — Building a local AI pipeline that normalizes multilingual input, infers action intent, and executes mega-prompts via the Llama CLI. 1. **Implementing graceful failure, JSON validation, and corpus testing** (23:52) — Mitigating LLM hallucinations by validating output against a strict schema, retrying failed generations at zero temperature, and running an evaluation corpus. 1. **Evaluating local LLM tradeoffs and user communication** (26:28) — Balancing the privacy benefits of local LLMs against high CPU usage, large binary sizes, and prompt ambiguity while communicating transparently with users. 1. **Live demonstration of local AI file renaming** (31:01) — Testing the local AI assistant live by translating filenames into German and attempting to replace spaces with emojis. ## Related Moments - [Preserving artificial intelligence history with localized open-source models](https://www.wearedevelopers.com/videos/100460-we-need-art-for-ai-safety) (from "We Need Art for AI Safety") - **Thumbnail URL:** https://customer-goifxnzhrq3d0d54.cloudflarestream.com/15463e191703199a7953f084b85c3ded/thumbnails/thumbnail.jpg?time=170443ms - [Summarizing developer experience and artificial intelligence companions](https://www.wearedevelopers.com/videos/884-forget-developer-platforms-think-developer-productivity) (from "Forget Developer Platforms, Think Developer Productivity!") - **Thumbnail URL:** https://customer-goifxnzhrq3d0d54.cloudflarestream.com/7e8924e91a0483cf0cf8990eb6122764/thumbnails/thumbnail.jpg?time=108855ms - [Designing AI applications defensively for inevitable failures](https://www.wearedevelopers.com/videos/100069-building-the-next-generation-of-ai-developer-tools) (from "Building the next generation of AI developer tools") - **Thumbnail URL:** https://customer-goifxnzhrq3d0d54.cloudflarestream.com/84605beada7b393b4f25327c680a42a7/thumbnails/thumbnail.jpg?time=163150ms - [Pushing back against predictable AI interfaces](https://www.wearedevelopers.com/videos/2144-developers-communities-in-san-francisco-wearedevelopers-live-from-corgi-cafe) (from "Developers Communities in San Francisco - WeAreDevelopers LIVE from Corgi Cafe") - **Thumbnail URL:** https://customer-goifxnzhrq3d0d54.cloudflarestream.com/93ad7c226ac352800568891ed3be3408/thumbnails/thumbnail.jpg?time=260031ms - [Enhancing conversational intent through modern large language models](https://www.wearedevelopers.com/videos/1641-hello-jarvis-building-voice-interfaces-for-your-llms) (from "Hello JARVIS - Building Voice Interfaces for Your LLMS") - **Thumbnail URL:** https://customer-goifxnzhrq3d0d54.cloudflarestream.com/c4aeff930240fcea57dd85942e8145f3/thumbnails/thumbnail.jpg?time=86045ms - [How generative models disrupt traditional machine interaction](https://www.wearedevelopers.com/videos/844-enter-the-brave-new-world-of-genai-with-vector-search) (from "Enter the Brave New World of GenAI with Vector Search") - **Thumbnail URL:** https://customer-goifxnzhrq3d0d54.cloudflarestream.com/9603ab4cad02218daecd3265703f19a3/thumbnails/thumbnail.jpg?time=184865ms ## Related Articles - [WWC24 Talk - Scott Hanselman - AI: Superhero or Supervillain?](https://www.wearedevelopers.com/magazine/469-wwc24-talk-scott-hanselman-ai-superhero-or-supervillain) - [Dev Digest 137 - AI'm not sure about this](https://www.wearedevelopers.com/magazine/485-dev-digest-137-ai-m-not-sure-about-this) - [MLOps And AI Driven Development](https://www.wearedevelopers.com/magazine/82-mlops-and-ai-driven-development) - [What Are Large Language Models?](https://www.wearedevelopers.com/magazine/304-what-are-large-language-models) ## Related Jobs - [AI Software Engineer (Germany)](https://www.wearedevelopers.com/jobs/48317-ai-software-engineer-germany) at **Sunhat** - [Principal Software Engineer, Copilot Code Review](https://www.wearedevelopers.com/jobs/ext/4667602-principal-software-engineer-copilot-code-review) at **GitHub** - [Staff Software Engineer, Copilot Code Review](https://www.wearedevelopers.com/jobs/ext/4743636-staff-software-engineer-copilot-code-review) at **GitHub** - [AI & Machine Learning Engineer (all genders)](https://www.wearedevelopers.com/jobs/48217-ai-machine-learning-engineer-all-genders) at **msg** - [Machine Learning Engineer 2](https://www.wearedevelopers.com/jobs/ext/4256068-machine-learning-engineer-2) at **Twilio** - [Senior Data & AI Architect](https://www.wearedevelopers.com/jobs/48602-senior-data-ai-architect) at **reeeliance IM GmbH**