#ai #uxdesign #generativeai #productivity #llm #zefina #techinnovation #softwareengineering #sarvamai | Jagath Saradagi

Why are AI chats still linear? 🤯 We’ve all felt the "Chatbot Loop." You’re reading a long, complex AI response. You get a doubt in paragraph one, you ask a question, and the answer appears... at the very bottom. Then comes the scroll: Scroll down for the clarification. Scroll back up to find your place. Repeat. 🔄 It breaks your cognitive flow and wastes your time. I built Zefina to fix this. I call it the "No-Nav" Experience. Instead of forcing you to the bottom, Zefina lets the AI answer right next to your doubt. It dynamically creates space inside the message, branches the conversation, and "slices" the context so the AI stays hyper-focused on exactly what you’re looking at. No more losing your place. No more endless scrolling. Just a recursive knowledge tree that grows with your curiosity. 🌳✨ 💡 A common real-life pain point this solves: "Imagine you’ve just finished studying an entire subject unit using an AI. You asked several follow-up questions to clarify complex parts. A week later, you return to revise. You remember asking for a simpler explanation of a specific paragraph, but that explanation is now buried at the bottom of a massive chat. To find it, you have to scroll through screens of text, find the answer, and then jump back up to see the original context. With Zefina, those clarifications are pinned exactly where they belong—right next to the source text. Revision becomes a seamless journey." ⚙️ Technical Note: Zefina is built with React and Framer Motion. The recursive brain is powered by Groq (Llama 3.3 70B), optimized with custom context-slicing logic for precise intra-message threading. I’m looking for feedback from fellow builders and researchers—how do you handle long AI conversations today? Would this recursive approach change how you work? Let’s discuss! 👇 I'm curious to get the perspective of the experts building the future of LLM interfaces. What do Andrej Karpathy , Logan Kilpatrick , Wulfie Bain or Amjad Masad think about moving away from linear scroll toward recursive knowledge trees? I'd also love to hear from the local leaders at Vivek Raghavan and Pratyush Kumar —on how this 'context-slicing' could optimize Indic LLM workflows. #AI #UXDesign #GenerativeAI #Productivity #LLM #Zefina #TechInnovation #SoftwareEngineering #SarvamAI

linkedin.com

Recursive discussions, anchored directly inside AI-generated Markdown responses.

Traditional AI chat interfaces force users to ask follow-up questions at the very bottom of a conversation—even when their doubt concerns one particular sentence in a long response.

@zefina/inline-discussions solves this by placing discussions where the question originates. Users can select any passage in an AI-generated Markdown response, open an inline input at that exact sentence, and create nested, recursive discussions without modifying or breaking the original Markdown response.

• 🎯 Precise Text Anchoring: Maps browser DOM selections back to exact raw Markdown source offsets using AST character maps.
• 🌳 Sibling-Block AST Injection: Splits Markdown paragraphs into valid sibling blocks at anchor boundaries. Original Markdown source text remains completely immutable.
• 🔄 Recursive Threading: Supports infinite nesting of discussions inside discussions (discussions branching from discussions).
• 🧩 Zero-Host Lock-In: Decoupled from any database, AI model, or backend. The…