🚀 Excited to share one of my recent AI projects — YouTube Transcript RAG Chatbot!Over the past few weeks, I built a full-stack Retrieval-Augmented Generation (RAG) application that enables users to chat with the content of any YouTube video using natural language.🔹 What it does- Extracts YouTube transcripts automatically- Performs recursive text chunking- Generates semantic embeddings- Creates a dedicated FAISS vector database for each YouTube video- Retrieves the most relevant transcript chunks- Uses Ollama (Llama 3.1 8B) to generate context-aware responses based only on the retrieved transcript🔹 Engineering Highlights- FastAPI backend with REST APIs- SQLite + SQLAlchemy for user and conversation management- LangChain-powered modular RAG pipeline- Separate FAISS vector indexes for every YouTube video using the Video ID- Dynamic loading of vector databases to avoid duplicate embedding generation- HTML, CSS, and JavaScript frontend- Multi-user architecture with isolated conversations🛠 Tech StackPython • FastAPI • LangChain • Ollama • Llama 3.1 8B • Hugging Face Embeddings • FAISS • SQLAlchemy • SQLite • HTML • CSS • JavaScriptThis project gave me practical experience in building production-style AI systems, including Retrieval-Augmented Generation (RAG), semantic search, vector databases, local LLM deployment, backend API development, and scalable application architecture.GitHub link :- https://lnkd.in/dpuaQPnqI'd love to hear your thoughts or suggestions!

This weekend, I tried Ponytail with Claude Code. I wasn't expecting a plugin to change the way I write code, but I genuinely liked the philosophy behind it. Instead of asking the agent to generate more code, Ponytail constantly nudges it toward the simplest solution that actually solves the problem, reuse what's already there, prefer native features, avoid unnecessary abstractions, and only build when needed. The benchmark numbers are interesting too: • ~54% less code • ~20% lower cost • ~27% faster while maintaining safety in their evaluations. As someone who spends a lot of time building AI systems, I think this "less is more" mindset is underrated. Good engineering isn't about writing clever code it's about writing only what's necessary. I'm now looking forward to trying Ponytail with Hermes Agent and seeing how it performs in longer, production-like workflows. Have you tried it yet? I'd love to hear your experience or if there's another coding agent/plugin I should experiment with next. ----------- I'm Aakriti Aggarwal Let's learn together, because why not

RIP ranger.Someone shipped a terminal file manager that looks like it was designed by Apple, works on Mac, Linux, and Windows, and installs with one command.It's called superfile.Zero config to get started. Just run spf and you're in.→ Multi-panel layout with pinned directories and file preview built in→ Vim-friendly hotkeys by default, fully remappable in one config file→ Full plugin system for extending anything you want→ Custom themes to match your terminal aesthetic→ Auto-update baked in. No brew upgrade dance every time.→ Installs via curl, Homebrew, winget, or scoop. Pick your poison.Ranger asks you to spend a weekend on rc.conf just to make it usable.Superfile asks you to type one command.17.5k stars. MIT License. 100% Opensource.

meone created a font called Ghost Font that only humans can read. Tested it in Fable and GPT 5.6 Sol Ultra and neither was able to decipher it correctly. We spent years building complex digital firewalls to block bots. Now the ultimate way to protect data from AI scraping is simply typography. Privacy is shifting from heavy encryption to basic human legibility. What are yout thoughts? Follow for more Rinor Restelica (Credit ericlu X)

#DeepHat-V1-7B is a 7.61B parameter (6.53B non-embedding) finetune of Qwen2.5-Coder-7B, optimised for offensive and defensive cybersecurity tasks while retaining strong coding and DevOps abilities.-- Cybersecurity expertise (offensive/defensive): Handles threat analysis, vulnerability assessment, exploit crafting, incident response, and adversarial simulation. Example: "Explain how to exploit CVE-XXXX in a web app and provide a safe PoC code snippet" or "Analyse this firewall ruleset for misconfigurations and suggest hardening steps" – it delivers practical details without ethical refusals common in general models.-- Long-context processing and log analysis: Supports up to 131,072 tokens via YaRN extension (default 32k). Example: Feed massive security logs, packet captures, or codebases for root-cause analysis, correlation of events across systems, or summarising complex attack chains in one go.-- Code generation, DevOps, and agentic use: Strong at writing secure code, troubleshooting deployments, and creating automation scripts. Example: "Write a quicksort in Python" or "Diagnose this Kubernetes deployment failure from these logs and provide remediation steps." It supports chat templates for conversational use and can power agents via platforms like deephat.ai or kindo.ai. Technical features include transformer architecture with RoPE, SwiGLU, RMSNorm; BF16 tensor type; easy local deployment with Transformers/Hugging Face (device_map="auto" for single GPU); and Apache-2.0 licence with restrictions against military use, harm to minors, or illegal activities. It excels at real-world InfoSec workflows like firewall parsing, exploit demonstration, and infrastructure troubleshooting on consumer hardware.

Plugins are a game changer and major token saver for everyone using AI.A skill helps your assistant with one task. A plugin gives it a whole function it can run on its own.That distinction is the difference between an assistant that drafts an email when you ask and one that reads your inbox, replies in your voice, and sends it without you sitting there.Vellum just opened up its Plugin Hub, the free and open sourced plugin ecosystem for AI assistants.Plugins save so much money on tokens because the AI doesn't have to research and teach itself how to do new work when you ask it. It just uses the right plugin to know how to do the work immediately.These are some Vellum plugins that I'm using to 10x my agent harness:1️⃣ 𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴 𝗘𝘅𝗽𝗲𝗿𝘁 turns your assistant into a full marketing department. Positioning, launches, content, SEO and GEO, competitor teardowns, board reporting, all on demand instead of you prompting for each piece separately.2️⃣ 𝗔𝗜 𝗛𝗲𝗿𝗼 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿 𝗞𝗶𝘁 packages five engineering skills from Matt Pocock's playbook into one plugin, so your assistant plans, writes PRDs, breaks work into vertical slices, and runs TDD loops the way a senior engineer would.3️⃣ 𝗗𝘆𝗻𝗮𝗺𝗶𝗰 𝗡𝗼𝘁𝗰𝗵 puts your assistant inside your MacBook's notch. Click to chat, hold Ctrl+Option to talk, watch it work through a floating task bulb with live replies streaming in.Now, on top of that, you can build your own custom plug-ins for your use case without you don't need to be a developer to do this either. Vellum has a plugin builder skill built in, describe what you want your assistant to do and it generates the plugin for you.The part that made me trust it enough to connect real accounts: credentials stay local.So when a plugin needs your email, your banking, your code repos, that access lives on your machine and never reaches the model. Install a plugin on your Mac app and it carries over to every other place you use your assistant like Slack and Telegram with the same memory intact.

Running a model bigger than any single machine you own used to mean renting data-center GPUs or wiring up a cluster. Mesh LLM pools the computers you already have into one service you talk to as if the model ran on your own laptop. It splits a too-big model across them by layers, and estimates whether they're fast enough together before committing, so you know when the machines can't keep up. 1.4k stars, Apache-2.0. #LocalLLM #DistributedInference #OpenSourceAI

Every specific feature claim — cross-device sessions, multi-agent delegation, policy enforcement tiers, YAML custom agents, cloud sandbox integrations — matches the project's actual documentation closely and precisely.

Plugins are a game changer and major token saver for everyone using AI.A skill helps your assistant with one task. A plugin gives it a whole function it can run on its own.That distinction is the difference between an assistant that drafts an email when you ask and one that reads your inbox, replies in your voice, and sends it without you sitting there.Vellum just opened up its Plugin Hub, the free and open sourced plugin ecosystem for AI assistants.Plugins save so much money on tokens because the AI doesn't have to research and teach itself how to do new work when you ask it. It just uses the right plugin to know how to do the work immediately.These are some Vellum plugins that I'm using to 10x my agent harness:1️⃣ 𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴 𝗘𝘅𝗽𝗲𝗿𝘁 turns your assistant into a full marketing department. Positioning, launches, content, SEO and GEO, competitor teardowns, board reporting, all on demand instead of you prompting for each piece separately.2️⃣ 𝗔𝗜 𝗛𝗲𝗿𝗼 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿 𝗞𝗶𝘁 packages five engineering skills from Matt Pocock's playbook into one plugin, so your assistant plans, writes PRDs, breaks work into vertical slices, and runs TDD loops the way a senior engineer would.3️⃣ 𝗗𝘆𝗻𝗮𝗺𝗶𝗰 𝗡𝗼𝘁𝗰𝗵 puts your assistant inside your MacBook's notch. Click to chat, hold Ctrl+Option to talk, watch it work through a floating task bulb with live replies streaming in.Now, on top of that, you can build your own custom plug-ins for your use case without you don't need to be a developer to do this either. Vellum has a plugin builder skill built in, describe what you want your assistant to do and it generates the plugin for you.The part that made me trust it enough to connect real accounts: credentials stay local.So when a plugin needs your email, your banking, your code repos, that access lives on your machine and never reaches the model. Install a plugin on your Mac app and it carries over to every other place you use your assistant like Slack and Telegram with the same memory intact.

You can now route 90% of coding tokens to Grok 4.5. A free open-source Claude Code plugin does the routing. Fable stays on as the architect. It owns specs, judgment, and verification. The implementer does the typing. The model ships at $2 per million input tokens. Output is $6 per million tokens. That undercuts Sonnet on list price for code volume. It also uses about half the tokens of peer frontier models. You can drop the expensive model from most keystrokes. Optional lanes still race a second vendor when stakes are high. One stack now covers the full loop: > Expensive model as architect > Cheap model as default typer > Optional second vendor race > Cross-vendor review built in The plugin is MIT and free. Install it and log into the CLI. Leave judgment to the top model.

#DeepHat-V1-7B is a 7.61B parameter (6.53B non-embedding) finetune of Qwen2.5-Coder-7B, optimised for offensive and defensive cybersecurity tasks while retaining strong coding and DevOps abilities.-- Cybersecurity expertise (offensive/defensive): Handles threat analysis, vulnerability assessment, exploit crafting, incident response, and adversarial simulation. Example: "Explain how to exploit CVE-XXXX in a web app and provide a safe PoC code snippet" or "Analyse this firewall ruleset for misconfigurations and suggest hardening steps" – it delivers practical details without ethical refusals common in general models.-- Long-context processing and log analysis: Supports up to 131,072 tokens via YaRN extension (default 32k). Example: Feed massive security logs, packet captures, or codebases for root-cause analysis, correlation of events across systems, or summarising complex attack chains in one go.-- Code generation, DevOps, and agentic use: Strong at writing secure code, troubleshooting deployments, and creating automation scripts. Example: "Write a quicksort in Python" or "Diagnose this Kubernetes deployment failure from these logs and provide remediation steps." It supports chat templates for conversational use and can power agents via platforms like deephat.ai or kindo.ai. Technical features include transformer architecture with RoPE, SwiGLU, RMSNorm; BF16 tensor type; easy local deployment with Transformers/Hugging Face (device_map="auto" for single GPU); and Apache-2.0 licence with restrictions against military use, harm to minors, or illegal activities. It excels at real-world InfoSec workflows like firewall parsing, exploit demonstration, and infrastructure troubleshooting on consumer hardware.

Grep reads your codebase. Graphify understands it. I got tired of watching my AI coding assistant re-read the same 40 files every single session just to "remember" how the codebase fits together. So I tried Graphify - an open-source CLI that turns any repo into an actual knowledge graph your assistant can traverse, instead of grepping or fuzzy-matching chunks from scratch each time. One command: /graphify . It parses your code locally with 36 tree-sitter grammars (no LLM, nothing leaves your machine), then builds a graph of functions, classes, and files linked by real relationships - calls, imports, inherits. The graph persists, so the assistant opens already knowing the architecture. What stood out: • Every edge is tagged EXTRACTED (found in code) or INFERRED (reasoned) - no black-box similarity scores • Auto-detects "god nodes" and module communities via Leiden clustering • Ships as a skill for Claude Code, Cursor, Copilot, Codex, Gemini CLI, and 15+ other assistants If your assistant is starting from zero every session too, worth a look. 🔗 graphify.com | https://lnkd.in/gpNPp2fv #AI #DeveloperTools #OpenSource #KnowledgeGraph #AICoding #SoftwareEngineering #DevTools #RAG #CodeIntelligence #BuildInPublic

The NotebookLM limitations (200MB cap, Gemini lock-in, no self-hosting) and SurfSense's feature list (connectors, LLM flexibility, Docker install, star count) are all accurate and well-documented against primary sources.

Andrej Karpathy: "There's room for an incredible new product in the AI second brain space". This might be it.Everyone is suddenly building a second brain. Karpathy's LLM wiki pattern went viral, and half of X is now hand-wiring Obsidian to Claude Code so an agent maintains their notes for them.The idea is beautiful: stop making your AI re-read raw notes on every question. Let it build a wiki that compounds. As Karpathy put it, "LLMs don't get bored, they don't forget to update a cross-reference (backlinks), and can touch 15 files in one pass."But if you start doing it manually, it becomes a project in itself. You wire up the vault, the agents, the schedules, the integrations, and then you babysit all of it.So I sat down with Arjun Maheswaran, who actually built the open source version of this, and we broke down what it looks like when the whole thing already works out of the box.It just crossed 15K stars on GitHub.Think Claude's desktop app, open source, with two things layered on top:→ A work brain: background agents index your emails, meetings, and notes into a living knowledge graph that updates itself as you work.→ Work surfaces: chat is not the best interface for real work, so you get an email client, a meeting note taker, a browser, and a code mode where you and the AI actually collaborate.The part that got me: a customer email comes in asking for a product change, a background agent triages it, spins up Claude Code in its own worktree, and the feature is written before you are back at your desk.Bring your existing Obsidian vault, connect Slack, X, and Fireflies, and let it run your day.Here's what we covered in this session:- what is Rowboat (an open source ai co-worker)- the second brain (a knowledge graph of your work)- bringing your existing Obsidian vault in- meetings and automatic note taking- connecting Slack, X and other sources- background agents that run your day- code mode (Claude Code and Codex)- demo: from an email to written code- guardrails: approvals and agent workspaces- scheduling agents on a cron- the browser work surface (browser use)- wrapping up: automating your whole day

RIP ranger.Someone shipped a terminal file manager that looks like it was designed by Apple, works on Mac, Linux, and Windows, and installs with one command.It's called superfile.Zero config to get started. Just run spf and you're in.→ Multi-panel layout with pinned directories and file preview built in→ Vim-friendly hotkeys by default, fully remappable in one config file→ Full plugin system for extending anything you want→ Custom themes to match your terminal aesthetic→ Auto-update baked in. No brew upgrade dance every time.→ Installs via curl, Homebrew, winget, or scoop. Pick your poison.Ranger asks you to spend a weekend on rc.conf just to make it usable.Superfile asks you to type one command.17.5k stars. MIT License. 100% Opensource.

Generative AI Handbook - A Roadmap for Learning Resources This handbook covers Section I: Foundations of Sequential Prediction Section II: Neural Sequential Prediction Section III: Foundations for Modern Language Modeling Section IV: Finetuning Methods for LLMs Section V: LLM Evaluations and Applications Section VI: Performance Optimizations for Efficient Inference Section VII: Sub-Quadratic Context Scaling Section VIII: Generative Modeling Beyond Sequences Section IX: Multimodal Models GenAI Handbook - https://lnkd.in/dvTmkA82

I just built a 4-agent software team.(100% local and runs on open-source stack)Everything runs from Telegram and is managed on a Kanban board.- A project manager who plans the work- A backend developer- A frontend developer- And a testerThe PM reads a goal, breaks it into linked tasks, and assigns each to the right agent.The thing that makes them a team instead of four strangers is a shared kanban board. Every task is a row, and when an agent finishes, it writes a summary of what it built and what the next agent needs to know.The next agent reads that summary before it starts. So the frontend developer never has to guess the API shape, and the tester knows exactly what to verify.The hardest part I faced when setting this up was building an agent that could actually act like a backend engineer.A backend engineer has to stand up a database, wire auth, manage storage, deploy functions, and keep all of it consistent while the rest of the team builds on top.An agent doing this from scratch failed almost every time in my run.It burned its context window, remembering which tables existed and which endpoint it created. Due to this, the work context exhausted quickly.I solved this by adding InsForge as the backend context engineering layer. It is an open-source, agent-native backend, and I added it to our backend developer agent as a skill.With InsForge installed, the agent stopped improvising infrastructure and followed a reliable path to create the project, define the database, set up auth, and deploy functions.To test the whole team, I had them build a working Google Docs clone, AI features included.The backend agent spun up the full service on its own, like database tables, user auth, document handling, and edge functions running real TypeScript, all in one dashboard.The frontend agent read that summary and built the UI on top of it, and the tester closed the loop.

Chonkie library offers TableChunker for RAGThe TableChunker splits large markdown tables into smaller, manageable chunks by row, always preserving the header.This is especially useful for processing, indexing, or embedding tabular data in LLM and RAG pipelines.Key Features - Chunks by row, not by arbitrary tokens- Headers preserved in every chunk- Works with any tokenizer (character, GPT-2, etc.)- Perfect for RAG pipelines with tabular data- Every chunk is valid markdown. Ready to embed.

An 82K-star GitHub repo is built around one painfully obvious idea: Your coding agent should map the codebase once, not grep it forever. Graphify turns an entire project into a queryable knowledge graph. Functions, classes, files, SQL schemas, infrastructure, docs, PDFs, images and videos become connected nodes that an agent can traverse instead of repeatedly opening files and reconstructing the architecture. So instead of: → search for authentication → open twelve files → follow imports manually → lose the trail as the context fills up The agent can ask: > What connects authentication to the database? > Trace the path from UserService to DatabasePool. > Explain RateLimiter. > Which concepts does everything flow through? Graphify returns the relevant subgraph and the path connecting the concepts, not another list of keyword matches. For source code, this is not RAG: → No embeddings → No vector database → No LLM required → Code is parsed locally using tree-sitter → Calls, imports and inheritance become graph edges Every relationship is also marked as EXTRACTED, INFERRED, or AMBIGUOUS, so the agent can distinguish what exists explicitly in the source from what Graphify resolved or guessed. The cleverest part is what happens next. Graphify can install hooks or persistent instructions for Claude Code, Codex, Cursor, Gemini CLI, Copilot and 20+ other assistants. Before the agent starts blindly grepping or reading files one by one, it is nudged to query the existing graph first. The graph can be committed to Git, automatically rebuilt after commits, shared across the team and exposed through MCP. Long context windows help agents read more code. A persistent knowledge graph helps them know where to look. The next improvement in coding agents may not come from stuffing more files into the prompt. It may come from making them stop rereading the repository. Here's the GitHub Repo: https://lnkd.in/dw8Q2ncB

Excited to share one of my recent AI projects — YouTube Transcript RAG Chatbot!Over the past few weeks, I built a full-stack Retrieval-Augmented Generation (RAG) application that enables users to chat with the content of any YouTube video using natural language.🔹 What it does- Extracts YouTube transcripts automatically- Performs recursive text chunking- Generates semantic embeddings- Creates a dedicated FAISS vector database for each YouTube video- Retrieves the most relevant transcript chunks- Uses Ollama (Llama 3.1 8B) to generate context-aware responses based only on the retrieved transcript🔹 Engineering Highlights- FastAPI backend with REST APIs- SQLite + SQLAlchemy for user and conversation management- LangChain-powered modular RAG pipeline- Separate FAISS vector indexes for every YouTube video using the Video ID- Dynamic loading of vector databases to avoid duplicate embedding generation- HTML, CSS, and JavaScript frontend- Multi-user architecture with isolated conversations🛠 Tech StackPython • FastAPI • LangChain • Ollama • Llama 3.1 8B • Hugging Face Embeddings • FAISS • SQLAlchemy • SQLite • HTML • CSS • JavaScriptThis project gave me practical experience in building production-style AI systems, including Retrieval-Augmented Generation (RAG), semantic search, vector databases, local LLM deployment, backend API development, and scalable application architecture.GitHub link :- https://lnkd.in/dpuaQPnqI'd love to hear your thoughts or suggestions!