We made our Ai agents 2x faster, 3x Cheaper with BAML ! скачать в хорошем качестве

We made our Ai agents 2x faster, 3x Cheaper with BAML !

How We Migrated to BAML and Made Our Agents 2x Faster In this video, I walk through how we migrated Flowlly's agent infrastructure to BAML (Boundary Abstract Modeling Language) and achieved significant performance improvements while solving some major headaches with LLM function calling. 🔗 Blog Post: https://web.flowlly.com/rahuls/baml-m... 🔗 BAML Documentation: https://docs.boundaryml.com/home 🔗 Flowlly: https://flowlly.com --- What we cover: What We're Building at Flowlly Understanding the Agent Loop Problem 1: Incompatible Schemas Problem 2: Invisible Prompt Injection Problem 3: Parallel Execution Bottleneck Problem 4: Model Lock-In Enter BAML Code Examples --- Resources: 📖 BAML Docs: https://docs.boundaryml.com/home 🔧 VSCode Extension: https://marketplace.visualstudio.com/... Provider Documentation: 🔷 Gemini Function Calling: https://ai.google.dev/gemini-api/docs... 🟢 OpenAI Function Calling: https://platform.openai.com/docs/guid... 🟣 Claude Tool Use: https://platform.claude.com/docs/en/a... --- Key takeaways: ✅ BAML abstracts away provider differences—write once, swap models anytime ✅ Schema-Aligned Parsing handles broken JSON without retries ✅ 92% accuracy with SAP on GPT-3.5 vs 87% with GPT-4o native function calling ✅ Full prompt visibility—see exactly what goes to the model ✅ Works with any model, even ones without native function calling support --- If you're building AI agents and struggling with latency, provider lock-in, or function calling complexity, this one's for you. 👍 Like if you found this helpful 🔔 Subscribe for more engineering deep dives 💬 Drop a comment if you have questions! --- #AI #LLM #BAML #Agents #Engineering #OpenAI #Claude #GPT #FunctionCalling #Python #TypeScript #Infrastructure #Flowlly #constructiontech