Built an Advanced Personal Assistant from scratch. Here's what it actually does. Started with a blank Next.js project and a FastAPI skeleton. The result is Ava — an AI assistant that reasons, remembers, and acts across sessions. The stack: Next.js 16 · FastAPI · SQLite · Groq API · Python Groq handles inference at blazing speed. Everything else — memory, plugins, sessions, file operations — runs on your own machine. ● Agentic tool calling: The LLM doesn't just respond — it decides. Every message goes through an orchestration loop that determines whether to answer directly or invoke a tool. Weather, time zones, calculations, web search, GitHub stats, crypto prices — all fire as live tool calls with transparent execution blocks in the UI. ● Multi-model fallback cascade: If the primary model hits rate limits, the system silently falls back through a chain of models without breaking the conversation. The user never sees an error. ● Code execution: Ava writes Python, runs it in a sandboxed subprocess, reads the output, fixes errors, and iterates — all in a single turn. The full execution trace is visible inline . ● Persistent memory: After every conversation, a background extraction pass pulls facts, preferences, and events into a structured vault. Location, tech stack, habits — remembered across sessions without any manual tagging. ● Voice and Vision: Push-to-talk via MediaRecorder piped to Groq Whisper for transcription. Image upload routes to a vision model for analysis, OCR, and structured extraction. ● Dynamic plugin system: Install and uninstall tools at runtime. Register a custom skill by uploading a markdown file — the parser extracts the schema and makes it callable immediately, no backend changes required. ● Session archive: Every conversation is stored and browsable. Restore any past session back into the live chat with one click. The hardest parts were never the features themselves. They were the details — preventing tool call JSON from truncating mid-generation, stripping internal reasoning tokens before they reach the UI, making a free tier feel unlimited through intelligent model routing. The gap between a working demo and a reliable product is where most AI projects fall apart. This one doesn't. Happy to go deep on any part of the architecture in the comments. #llm #nextjs #fastapi #python #ai #groq #softwaredevelopment #webdevelopment

Building a smarter search with RAG I spent the last 24 hours building The Atheneum, a search tool that uses Retrieval-Augmented Generation (RAG) to help find books based on context rather than just titles. The Goal: Most book sites fail if you don't know the exact name. I wanted to see if I could use vector embeddings to let users search by "feeling" or "intent." The Technical Breakdown: Scraping: Built a custom Python script (BeautifulSoup) to pull and clean data. Vector Store: Integrated ChromaDB to store book descriptions as embeddings. Intelligence: Used a RAG pipeline to match natural language queries to the most relevant books in the database. Frontend: Built a clean React interface using Tailwind CSS and Framer Motion to handle the AI's "Librarian Console" responses. Why this mattered to me: Instead of just building a basic CRUD app, I had to figure out how to handle "dirty" scraped data, manage vector similarity, and keep the UI responsive while the AI processes a request. It was a great challenge in combining data engineering with a functional user experience. #BuildInPublic #Python #Django #React #AI #RAG #SoftwareEngineering #DeepLearning #OpenSource

Day 62 of 100 Days of AI — 🔗 Everything is Connected Today was one of those days where everything clicks into place. Every piece of the AI Newsletter is now talking to each other. Cloudflare Worker scrapes the sources on schedule → sends raw content to the Python backend → LangChain agents curate, synthesize, and rewrite → Jinja2 renders the final HTML email → Resend delivers it to subscribers. Next.js handles the frontend — clean landing page, subscribe form, all connected to the backend. OpenRouter powering the LLM calls. LangGraph orchestrating the agent pipeline. FastAPI holding everything together in the middle. Seven different services. One clean flow. Zero manual steps. The part that felt most satisfying — watching a raw scraped article go all the way through the pipeline and come out the other end as a polished newsletter section. Automatically. No human involved. That's what 62 days of building toward this feels like. Tomorrow — production deployment. Real subscribers. Live. Next: Production deployment — the newsletter goes live. #100DaysOfAI #BuildInPublic #LangChain #LangGraph #CloudflareWorkers #FastAPI #NextJS #AIEngineering #Newsletter #OpenRouter #Python

Anthropic accidentally leaked Claude Code's entire 500,000+ line TypeScript source code through an npm package that included a development source map file. This ironic leak from the "safety-first" company revealed that their AI coding assistant is essentially sophisticated prompt engineering rather than revolutionary technology, along with unreleased features and anti-competitive tactics. A few points: 1. The leak happened through a 57MB source map file accidentally included in npm package version 2.1.88, possibly due to a bug in Bun.js 2. Claude Code is built with 11 processing steps but relies heavily on hard-coded prompts and guardrails rather than advanced AI architecture 3. Anthropic implemented "anti-distillation poison pills" - fake tools designed to sabotage competitors trying to copy Claude's outputs 4. "Undercover mode" instructs Claude to hide its AI identity in outputs, potentially to make AI-generated code appear human-written 5. The system uses basic regex patterns to detect user frustration through keywords like "damn" and "balls" 😂 6. Leaked code revealed unreleased features including "Buddy" (AI companion), "Chyus" (background agent), and references to future models 7. The open-source community quickly created "Claw Code" - a Python rewrite of the leaked TypeScript code

What if your journal could write back? 🌸  I built MindEase API ... a mood journaling backend where an AI therapist named Luna responds to how you feel. In under a second. No login. No setup. Just send an emoji + your thoughts… and Luna takes care of the rest. 💬 3 Simple Endpoints 1. Talk to Luna Send your emoji + message → Luna replies instantly using GROQ’s llama-3.1-8b-instant. The entry saves automatically. And if GROQ has a bad day? I built a graceful fallback. No crashes. Ever. 2. Your Mood History GET all your past entries — newest first. Every conversation with Luna. One call. 3. Luna’s Weekly Letter This one hits different. Luna reads your entire week. Spots your dominant emoji. Calculates your streak. Then writes you a personal letter — starting with “Dear friend,” and ending with “— Luna 🌿” One message isn’t a week. She needs at least 2 entries. 🔧 Under the Hood Django REST + DRF serializers for clean validation GROQ llama-3.1-8b-instant — fast, free, and incredibly warm drf-spectacular for auto Swagger + ReDoc Custom prompts — Luna writes differently in chats vs weekly letters DB indexing on user_id because slow queries are a choice GROQ key in Railway env vars (never in code) 🔗 Try It and check out the source code if you want https://lnkd.in/d8fSmszF https://lnkd.in/dDAz2kMB API Diagram in the first comment 👇 Next post .... the Flutter app I built on top of this API 🔥 Stay tuned. #Django #DjangoRestFramework #Python #GROQ #GenerativeAI #BuildInPublic #Flutter #FlutterDev #BackendDevelopment #APIDesign #Portfolio #DevCommunity #100DaysOfCode #MindEase

🤖 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭𝐬 𝐝𝐨𝐧'𝐭 𝐡𝐚𝐥𝐥𝐮𝐜𝐢𝐧𝐚𝐭𝐞 𝐢𝐧 𝐩𝐚𝐫𝐚𝐠𝐫𝐚𝐩𝐡𝐬 𝐚𝐧𝐲𝐦𝐨𝐫𝐞. 👻 𝐓𝐡𝐞𝐲 𝐡𝐚𝐥𝐥𝐮𝐜𝐢𝐧𝐚𝐭𝐞 𝐢𝐧 𝐉𝐒𝐎𝐍. Last week, I spent 2 hours debugging why my agent kept failing a tool call. GPT-5.2 output v1: "score": "42" GPT-5.2 output v2: "score": 42 Switching keys or types can really throw off your whole workflow—that's just the tame example. If you're working in the AI world, you’re probably all too familiar with the real challenges: comparing two LLM function calls side by side, checking what changed in your RAG retrieval payload, debugging why your API v2 returns an extra nested field, or reviewing agent memory before and after running a tool. We tend to handle these tasks the same old way—opening two VS Code tabs, squinting, scrolling, and often missing that one crucial detail. That’s why I created the tool I really needed: 𝐉𝐒𝐎𝐍 𝐃𝐢𝐟𝐟 𝐕𝐢𝐞𝐰𝐞𝐫. It’s open source, instant, and accessible without any login. Just paste JSON A and JSON B, hit compare, and it shows you exactly what’s new in green, what’s been removed in red, what’s changed in yellow, and any type changes in orange (this last one has saved me multiple times!). It’s exactly like in the screenshot—side by side, clear, distraction-free. I built this because JSON is the most reliable way we have to understand the contracts between models, tools, and APIs in agentic systems—but we’ve lacked good tools to see what actually changed. It's free, open source, and works right in your browser. If you’re building agents, LLM apps, or working with APIs every day, give it a try—you'll wonder how you ever managed without it. Think of how many hours you might have lost last month to a missing comma or a confusing string versus a number. Find the link in the first comment, or comment "𝐃𝐈𝐅𝐅" and I'll DM it to you. #buildinpublic #opensource #aiagents #llm #developers #javascript #python #claudecode #anthropic

🚨BREAKING: Anthropic just accidentally leaked the ENTIRE source code of Claude Code Yesterday (March 31), a 59.8 MB JavaScript source map slipped into the official @anthropic-ai/claude-code npm package. Inside? → 512,000+ lines of clean, unobfuscated TypeScript → The full agentic harness that turns Claude into a high-agency coding beast → Complete system prompts, tool implementations, memory architecture, and reasoning loops → Sneak peeks at unreleased features (KAIROS always-on agent, undercover mode, Tamagotchi-style UI elements, and more) This wasn’t a hack. No weights, no training data, no customer info leaked. It was a classic human error in packaging they simply forgot to exclude the .map files. But the internet moves fast. Security researchers grabbed it, mirrors popped up on GitHub, DMCA takedowns started flying… and then someone did the funniest thing: they used AI to rewrite the whole thing clean-room in Python (already sitting at tens of thousands of stars). This is the second major slip-up from Anthropic in a week (after the Claude Mythos data store incident). Why this actually matters: We now have a rare public window into how one of the top labs builds production-grade AI agents The gap between “chatbot” and “true coding teammate” just got a lot more transparent Every AI engineer and startup building agents is studying this right now The moat isn’t disappearing the models are still safe but the agent architecture playbook just got community-sourced overnight. This is what 2026 looks like: one packaging mistake and suddenly the entire industry levels up. What do you think? Is this a massive L for Anthropic or the ultimate accidental gift to the AI community? Drop your hot take below #AI #Claude #Anthropic #ClaudeCode #AgenticAI #DevTools #TechNews #OpenSource

Anthropic accidentally leaked Claude Code's entire source code.  512,000 lines. 1,900 TypeScript files. Forked 41,500 times before anyone could stop it. And what the internet found inside is wild. Claude Code detects when you're frustrated with it. Not with ML. Not with a fine-tuned model. With regex. Pattern matching for profanity and emotionally charged language. Pragmatic? Yes. Hilarious? Also yes. There's an "undercover mode." Claude Code can be instructed to hide internal codenames ("Capybara", "Tengu") and never admit it's Claude Code. One-way enforcement. No force-off switch. There's a secret unreleased agent called KAIROS. Always-on background agent with GitHub webhooks, cron jobs, and "nightly memory distillation." Feature-gated. Not shipped yet. And fake tools are injected into the system prompt deliberately to poison any competitor trying to distill the model's behavior from traffic. How did it leak? A 59.8MB source map file was accidentally bundled into the npm package for Claude Code v2.1.88. Source maps are meant for debugging. They point back to the original TypeScript. Anthropic confirmed: human error. No customer data exposed. Just their entire codebase. The irony: the world's most capable AI coding tool leaked because of a shipping mistake any developer could make. Anthropic said it best: "human error." Even the companies building AI aren't immune to very human problems. #anthropic #claudecode #ai #sourcecode #aitools #javascript #npm #softwareengineering #llm #devsecops

I just finished a cool new project called Noctra! 🚀 Reading through hundreds of messy customer complaints takes way too long. So, I spent the last few weeks building an AI app that does it for you. It reads through raw feedback and instantly gives you a short summary, points out the top 3 biggest problems, and tells you exactly how to fix them. 🛠️ How I built it (The Process): I broke the project down into three simple steps: The Data Pipeline: I set up a Python backend that grabs all the customer complaints from a JSON file (which can easily be upgraded to a real PostgreSQL database later!). The AI Brain: Using FastAPI, I connected my code to Google's Gemini AI. Instead of asking the AI one giant question, I chained the prompts: the AI first reads everything to make a summary, then it rereads that summary to find the biggest bugs, and finally, it comes up with specific solutions. The User Interface: Finally, I built a React front-end to grab those answers from the AI and display them on the screen. 🧠 The Challenge: I wanted to skip the boring "standard dashboard" look, so the biggest challenge was styling the data to look like glowing, floating glass using pure CSS animations! 💡 What I Learned: Connecting different tools together sounds scary, but if you take it step-by-step (Data -> Backend AI -> Frontend UI), it's actually very straightforward! #WebDevelopment #ReactJS #FastAPI #Python #AI

I built an 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁 that reads your code and decides, 𝗼𝗻 𝗶𝘁𝘀 𝗼𝘄𝗻, what kind of help it needs. 𝗖𝗼𝗱𝗲𝗦𝗲𝗻𝘀𝗲 - 𝗔𝗻 𝗔𝗜-𝗽𝗼𝘄𝗲𝗿𝗲𝗱 𝗰𝗼𝗱𝗲 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁 𝗮𝗴𝗲𝗻𝘁 that accepts a code snippet and a natural language instruction, reasons about the intent, and autonomously selects and runs only the relevant analysis tools.   The six tools it can call:  • Bug Detector  • Security Vulnerability Checker  • Code Quality Analyzer  • Code Explainer  • Unit Test Generator  • Refactor Suggester Each tool runs on a focused 𝘀𝘆𝘀𝘁𝗲𝗺 𝗽𝗿𝗼𝗺𝗽𝘁 engineered to return strict structured JSON - making the output predictable, parseable, and renderable as distinct UI components. 𝗛𝗼𝘄 𝗱𝗼𝗲𝘀 𝗶𝘁 𝘄𝗼𝗿𝗸? A 𝗟𝗮𝗻𝗴𝗖𝗵𝗮𝗶𝗻 agent powered by 𝗚𝗼𝗼𝗴𝗹𝗲 𝗚𝗲𝗺𝗶𝗻𝗶 reads both the code and your instruction, reasons about your intent, and selects only the relevant tools from a pool of six. It does not run all six blindly. It picks what makes sense, executes those tools, observes the results, and decides if anything else is needed before giving you a final answer. That decision-making loop is what makes it an actual agent. 𝗧𝗲𝗰𝗵 𝘀𝘁𝗮𝗰𝗸: Python, FastAPI, LangChain, Google Gemini API, React.js, Tailwind CSS - deployed on Render and Vercel. Live link and GitHub Repo in the comments. #AI #LangChain #GenerativeAI #AgenticAI #Python #FastAPI #React #FullStackDevelopment #LLM #MachineLearning #OpenToWork