标签：#AI

I built AIPDFKit because I kept running into the same friction: needing to do something simple with a PDF -- redact some sensitive info, pull out a table, or convert a document to Markdown -- and every tool either required an account, put the good stuff behind a paywall, or made me wonder what was happening to my files afterward. PDFKit is my answer to that. PDFKit -- Free AI-Powered PDF Tools PDFKit is a free, browser-based PDF utility suite powered by AI, built for developers and technical professionals who need fast, reliable document processing without the friction of paid plans or mandatory accounts. Whether you're parsing data out of PDFs, sanitizing sensitive information, or converting documents into developer-friendly formats, PDFKit gets the job done in seconds. What it does AI-assisted PII redaction -- automatically detect and mask emails, phone numbers, names, and more Table extraction to Excel -- pull structured data out of PDFs without copying and pasting PDF to Markdown conversion -- especially useful for feeding document content into LLMs or RAG pipelines These aren't just format converters. The AI layer means the output is clean, structured, and actually ready to use. Privacy first No account creation required. PDFKit stores no user data and automatically deletes all uploaded files after one hour. For developers handling client documents or sensitive data pipelines, this is a meaningful differentiator over SaaS tools that retain files indefinitely. Who it's for Developers preprocessing PDFs before feeding them into RAG pipelines Anyone automating document workflows People who need to quickly extract structured data without spinning up a Python script Anyone dealing with sensitive documents who can't afford to have files sitting on someone else's servers It's the kind of utility you bookmark and reach for constantly. Built to be fast, free, and frictionless. Check it out: https://www.aipdfkit.com/ Would love to hear what features you'd find most usefu

This is the eighth and final post in a series on the false assumptions teams make when building with generative AI. The series began with the observation that the trough of disillusionment for AI-assisted development has arrived — not because AI is useless, but because eight false assumptions made the trough inevitable. This post covers the last assumption and closes the series. The Fallacy "If one AI agent gives us a 10x boost, ten agents will give us 100x." Why it's tempting The arithmetic feels irresistible. One agent generates code for the backend. Another generates the frontend. A third writes tests. A fourth handles database migrations. A fifth generates documentation. Each agent works in parallel. No meetings, waiting or coordination overhead. Pure throughput. Leadership sees the potential: a five-person team with fifty agents has the output of a fifty-person team at the cost of a five-person team plus API credits. The scaling is linear. The economics are transformational. And the early results confirm it. Each agent, working on its own, produces impressive output. The backend agent generates Go code. The frontend agent generates React components. The test agent generates test suites. Each agent, in isolation, looks like a 10x developer. Why it's wrong You've seen this problem before. It has a name. It's called distributed systems. A distributed system is a collection of independent actors that must coordinate to produce a coherent result. Each actor makes decisions locally. The system's correctness depends on those local decisions being compatible globally. When they aren't, you get inconsistency, conflicts, data corruption, and cascading failures. AI agents working on the same codebase are a distributed system. Each agent makes decisions — variable names, error handling strategies, retry policies, data formats, abstraction levels, dependency choices. Each decision is made locally, in the context of one prompt, one file, one task. No agent sees the full pict

Hello, my name is Greg. About six months ago I started using AI chatbots like ChatGPT and Claude at work for small tasks — proofreading emails, summarizing meeting notes, that kind of thing. But also some technical stuff too. One thing that really came in handy was analyzing packet captures from Wireshark. I work with VoIP phone systems, and when things go wrong, feeding a PCAP file into an AI chatbot speeds up the troubleshooting process dramatically. Before long I was asking AI to write code. First simple HTML pages, then Python, then C#. I was amazed by the results. These weren't big projects — just small experiments — but they came to life in minutes instead of days. I found out there was already a term for this: vibe coding . Perfect, I thought. I made project after project and wanted to share the excitement with other people who were surely doing the same thing. I created a free learning website, published a book on Kindle Unlimited, and went looking for a community. I landed on Reddit. There were already vibe coding subreddits. I thought — this is great, I've found my people. Then reality hit. These communities had "vibe coding" in the name, but they weren't exactly vibe coding friendly. The term had already been claimed by people focused on monetizing their creations fast, with little interest in actually learning to code. That created a massive anti-vibe-coding crowd on the other side, and honestly there was an all-out war going on between them. Not really the place for someone just looking to share cool stuff they made. I came to a realization: I wasn't really a vibe coder — at least not the kind people were arguing about. I wasn't in it for the money. I was in it for the fun. I didn't mind learning programming concepts along the way. I wasn't trying to sell anything or launch a startup. I just liked making things and solving problems. So I retreated and regrouped. That's when I found a better description: Play-First Programmers . People who start by playi

We are exploring an experimental blockchain mechanism called "Proof of Weather" In the world of blockchain, various methods are used to achieve network consensus. The most well-known is Bitcoin’s Proof of Work (PoW). While PoW is an excellent mechanism, it has one major drawback. It consumes an enormous amount of electricity. At one point, I found myself wondering: Does blockchain really require such vast computational resources? Isn’t there something else that’s needed? This led to the creation of Dawn, the experimental cryptocurrency project I am developing, and an experimental blockchain mechanism called Proof of Weather. In this article, I will discuss: Why I decided to use weather How Proof of Weather works Security considerations Implementation in Rust How Does Proof of Work Work? Proof of Work is often explained as a mechanism where computers compete against each other in computational tasks. However, one important property of PoW is that it produces outcomes that are difficult to predict in advance. Miners repeatedly perform massive amounts of hash calculations, and only those who happen to meet the conditions can generate a block. This unpredictability plays a role in determining who can produce the next block. However, this process consumes enormous amounts of electricity worldwide. So I wondered: Aren’t there already phenomena in nature that are difficult to predict? Why Weather? Proof of Weather utilizes weather data as that unpredictable element. Of course, weather forecasts exist. However, Temperatures several days in the future Atmospheric pressure at specific locations Precipitation Wind speed and other factors cannot be predicted with absolute certainty. In particular, when combining observations from multiple locations, it becomes even more difficult to accurately calculate future values in advance. In other words, meteorological observations have the potential to be used as A real-world information source where future values cannot be fully predic

Apple has been on its back foot, AI-wise, for the past few years. But in a strange way, playing from behind might not be such a bad move. At WWDC on Monday, Apple appears to be getting ready to reintroduce us to the new Siri. Again. As a reminder, we met the new Siri in […]

Google has just signed a $30 billion AI computing power deal with SpaceX.

This is a submission for the GitHub Finish-Up-A-Thon Challenge OpsPilot AI: Reviving an Unfinished AI-Powered Operations Platform with GitHub Copilot What I Built OpsPilot AI is an AI-powered operations assistant designed to help DevOps engineers, SREs, and operations teams investigate incidents, monitor service health, and gain actionable operational insights. The project originally started as a side project inspired by my experience working in production support and monitoring environments. I built an initial version to validate the idea but never fully completed it. The core concept was promising, but several important features and usability improvements were still missing. Through the GitHub Finish-Up-A-Thon Challenge, I revisited the project and transformed it into a much more complete and polished MVP. Key features include: AI-powered incident analysis Root cause investigation assistance MTTR analytics dashboard Service health monitoring Incident trend analysis Executive reporting insights Modern responsive user interface Demo Live Application GitHub Repository OpsPilot AI helps operations teams reduce investigation time and improve operational visibility through AI-powered workflows and analytics. The Comeback Story When I first started OpsPilot AI, it was mainly an experiment to explore how AI could assist operations teams during incident investigations. Although the foundation was built, the project was left unfinished because of limited time and competing priorities. The original version lacked: Incident analytics Meaningful operational insights Root cause investigation workflows Executive reporting capabilities A polished user experience For this challenge, I focused on completing the project and turning it into a usable MVP. What I Added AI Incident Analysis Enhanced the platform with AI-powered incident summaries and investigation assistance. Operations Analytics Added dashboards to track: Mean Time To Resolution (MTTR) Incident frequency Service health

Rails GuardDog: Advanced Security Scanner for Rails Introduction Today I'm excited to announce Rails GuardDog v0.1.0 — an open-source security scanner for Rails that goes beyond traditional tools like Brakeman. While Brakeman is excellent for catching basic Rails vulnerabilities, Rails GuardDog focuses on newer vulnerability classes that most tools miss: AI/LLM prompt injection, DoS/ReDoS patterns, supply chain attacks, and more. The Problem Modern Rails applications face new security challenges: AI/LLM Integration - How do you prevent prompt injection when integrating with ChatGPT, Claude, or Anthropic? ReDoS Attacks - Catastrophic backtracking in regex can bring down your app Supply Chain Attacks - Typosquatted gems that look like popular libraries IDOR Gaps - Objects accessible without proper authorization checks Advanced Secrets - Hardcoded API keys that Brakeman misses Rails GuardDog detects all of these. What is Rails GuardDog? Rails GuardDog is a lightweight gem that adds comprehensive security scanning directly to your Rails applications. 12 Security Checkers SQL Injection - String interpolation in queries XSS - Unescaped output in views CSRF - Disabled protection verification Mass Assignment - permit! vulnerabilities (fixes Brakeman #1942, #1918) Open Redirect - User input in redirects Hardcoded Secrets - API keys, tokens, passwords (always-on, fixes #1989) DoS/ReDoS - Unbounded queries, dangerous regex patterns IDOR - Object access without authorization AI/LLM Prompt Injection - User input flowing to LLMs Rate Limiting - Missing rack-attack configuration Supply Chain - Typosquatted gems using Levenshtein distance GraphQL - Missing field-level authorization Features 📊 Multiple report formats : Console, HTML, JSON 🔍 AST-based analysis : Uses parser gem for deep code understanding ⚡ Async support : Built-in Sidekiq integration 📈 Zero dependencies : Only requires parser and ast gems 🚀 Production-ready : Tested and battle-ready 📝 CWE/OWASP mappings : Every find

TL;DR Claude Code and the op CLI reorganized 690 credentials — four vaults, 390 items tagged, SSH agent configured — in one session. This is AI-native work: the agent operated the vault; the human set direction and approved via Touch ID. The CLI failed on 18 items with social-auth ( UNKNOWN field type) — hard failure, not graceful degradation; a real reliability blocker for team-scale use. The bug was filed from the terminal via the GitHub CLI in the same session it was found. If your password manager has a CLI, you already have everything needed to run this. I've been a 1Password user for years. Not in a conscious, intentional way — more in the way you use a good chair: it became part of how I work and I stopped thinking about it. That changed when I set up a new machine. I had to install 1Password, wire up the SSH agent, reconnect the CLI, re-authenticate everything. The process took longer than it should have because I'd never written down what I'd built. I'd only accumulated it. And somewhere in the middle of that setup, it hit me: I had 690 credentials in one flat vault — logins from jobs I'd left years ago sitting next to active API keys, personal bank accounts mixed with infrastructure credentials, demo user passwords alongside production secrets. The kind of accumulation that happens when a tool works well enough that you never stop to organize it. I'd been meaning to clean it up for a long time. I never did, because the job is exactly the kind of work that's too tedious to do manually and too important to skip: touch every item, make a judgment call, move it somewhere sensible, repeat 690 times. Then I realized: with Claude Code and the op CLI, this was now actually possible. Not assisted — the agent could do it. So I handed it the keys. What "AI-native" actually means here Quick context on timing: 1Password launched its SSH agent and CLI 2.0 in March 2022. Git commit signing via the vault came six months later. These are mature, stable features — not betas

Day 48 of building GoDavaii, and the toughest problem isn't the sheer volume of allopathic medicines or the complexity of their interactions. It's the invisible logic of 'Desi Ilaaj' - the home remedies and traditional practices deeply ingrained in Indian families for generations. When everyone knows the comfort and efficacy of 'haldi-doodh' (turmeric milk) for a cold, how does an AI health platform authentically verify and integrate that knowledge without replacing professional medical advice? This isn't just a cultural nod; it's a fundamental challenge for any health AI truly built for India. Global competitors like Epocrates or drugs.com, while excellent within their scope, are entirely English-centric and focused on Western allopathic data. They have no framework for the millions of people who search for health guidance in Hindi, Tamil, or Marathi, and whose first instinct for a cough might be a herbal concoction, not an over-the-counter syrup. The Unspoken Truth About India's Health Landscape For a vast majority of Indian families, health decisions often involve a blend of modern medicine and traditional wisdom. From specific herbs to dietary adjustments passed down through generations, these practices are effective for many minor ailments. Yet, in the digital health space, they're largely ignored. Why? Because the data is fragmented, often anecdotal, and doesn't fit neatly into structured pharmacological databases. It's a goldmine of practical health knowledge, but also a minefield for safety if not handled with care. My realization as Pururva Agarwal, 27-year-old founder of GoDavaii, was simple but profound: if we truly want to serve families coming online in their mother tongue, our AI needs to understand and interact with this context. This means going far beyond just translating English medical terms into 22+ Indian languages. It means building a knowledge graph that can intelligently cross-reference traditional remedies with known active compounds, potent

Six months ago, I sat down with a YouTube playlist, a blank notebook, and one goal: learn Python. What I did not expect was how hard it would be, not the Python itself, but figuring out how to actually learn it. I started with a YouTube playlist. Simple enough. Except nobody tells you what to do after you watch a video. Do you rewatch it? Take notes? Jump straight to code? I had no system. I'd watch a concept, feel like I understood it, open VS Code, and stare at a blank file. That's when I realized I had fallen into passive learning. And passive learning in the age of AI is a particularly dangerous trap, because it's so easy to confuse activity with progress. I could watch a video, feel good. I could ask Claude to explain a concept, feel good. I could even ask AI to write code, read it, nod along, and feel like I'd learned something. I hadn't. I'd just consumed. There's a difference. The real moment of honesty came when I was stuck on a coding problem. My instinct, everyone's instinct now is to open ChatGPT or Claude immediately. And I knew, sitting there with the cursor blinking, that if I did that every single time I got stuck, I was building nothing. My brain would never develop the muscle of working through problems. I would be someone who can prompt AI to code, not someone who can think in code. And in a world where AI can already write decent code, the person who can't think independently isn't valuable. They're replaceable. So I had to build a system that forced me to actually learn. After a lot of trial and failure, I landed on a 5-phase checklist that I wrote out by hand and kept next to my laptop. Phase 1: is what I call First Contact — watch one focused video, then write a summary purely from memory, then discuss it with an LLM not to get answers but to pressure-test what I thought I understood. Phase 2: is Deep Understanding — read a written source, write proper notes, map the concept visually, and list every edge case and exception I can find. Phase 3:

In the previous article, we explored how to build our first CRUD API using FastAPI. While our API worked correctly, there was one major problem. We were storing data inside Python lists, which exist only in memory. If you've ever wondered how applications like Instagram, LinkedIn, or ChatGPT remember information even after a server restart, the answer is simple: databases. In this article, we'll solve the problem of in-memory storage by connecting our FastAPI application to SQLite using SQLAlchemy. If you haven't read the previous post, check it out: FastAPI for AI Engineers - Part 2: Building Your First CRUD API Ananya S Ananya S Ananya S Follow Jun 1 FastAPI for AI Engineers - Part 2: Building Your First CRUD API # ai # backend # fastapi # python 7 reactions Comments Add Comment 4 min read By the end of this article, you'll understand: Why in-memory storage is a problem What SQLite is What SQLAlchemy is How ORM works How to create database tables using Python classes How to perform CRUD operations using a real database The Problem with In-Memory Storage Previously, our application stored students inside a Python list. students = [ { " id " : 1 , " name " : " Ananya " , " department " : " CSE " , " cgpa " : 8.9 } ] This worked for learning CRUD operations. However, consider what happens when the server restarts: FastAPI Server Stops ↓ Python Memory Cleared ↓ All Student Data Lost This is unacceptable in real-world applications. We need a place where data can survive application restarts. This is where databases come in. What is SQLite? SQLite is a lightweight relational database. Unlike MySQL or PostgreSQL, SQLite doesn't require a separate database server. Instead, everything is stored inside a single file. students.db Advantages of SQLite: No installation required Lightweight Easy to learn Perfect for local development Great for small projects For this article, we'll use SQLite. What is SQLAlchemy? Before SQLAlchemy, developers often wrote raw SQL queries. Exampl

Why retrieval-augmented generation has become the foundational pattern for building useful AI — and how it actually works. The Problem With Relying on LLMs Alone Large language models are impressive. They can write, reason, summarize, and explain across an enormous range of topics. But they have a hard boundary: their knowledge stops at their training cutoff. Anything that happened after that date, anything specific to your company, your codebase, or your documents — the model simply doesn't know it. The naive solution is to paste your data directly into the prompt. For short content, this works. But prompts have limits. A model can only process so much text at once, and even within that limit, quality degrades when you stuff too much context in. The model loses track of things buried in the middle, confuses similar passages, and starts guessing when it should be reading. RAG — Retrieval-Augmented Generation — solves this properly. Instead of sending everything to the model and hoping for the best, you send only what's actually relevant to the question being asked. The Core Idea The analogy that makes RAG click immediately: imagine a student sitting an open-book exam. They don't memorize the entire textbook. When they see a question, they flip to the right chapter, read the relevant section, and write their answer from what they just read. They're not guessing. They're grounding their answer in the source material. RAG does exactly this. When a user asks a question, the system finds the most relevant pieces of information from your data, hands those pieces to the LLM as context, and the model answers from that context alone. The result is accurate, grounded, and verifiable — you can point to exactly which source the answer came from. The process runs in two phases: ingestion, which prepares your data in advance, and retrieval, which happens at query time. Phase One: Ingestion Ingestion is the preparation step. Before any user asks anything, you process your data and

A blind test across 240 images and 10 professional designers just dropped. Ideogram 4.0 against Gemini 3.1, Grok Imagine, and FLUX.2 Max. The results are clean. Ideogram won typography in nearly half of every blind matchup. 47.9 percent. Next closest was Gemini at 30 percent. FLUX.2 and Grok sat around 15 percent each. On the question that actually matters to designers -- would I ship this -- Ideogram scored 3.55 out of 5. Gemini got 2.84. Nobody else cleared 3. That is a real lead in text rendering. The model was trained exclusively on structured JSON caption datasets, which means it understands composition and layout differently than models trained on alt-text scraped from the web. The JSON prompting is genuinely useful for automated pipelines. You can specify bounding boxes, color palettes, object positions. It is not just better at text. It is more controllable. I tested it. It works. The text in images is readable. That has been the white whale of AI image generation for two years and Ideogram 4.0 mostly solves it. But as an overall image model, it is just good. Competitive, not dominant. On busy, highly detailed scenes with specific counts and attributes, Ideogram scored 3.42. Gemini scored 3.37. That is a statistical tie. FLUX.2 scored 3.01 and Grok 2.82, which are worse, but the gap between the top two is noise. For general image quality, you are splitting hairs between Ideogram and Gemini. For photorealism, FLUX and Reve still lead. For artistic generation, Midjourney is Midjourney. The prompting behavior is interesting. Lean prompts won across the board. Long, over-specified prompts lost. The model was trained on structured data, so it wants structure, not paragraphs. "A poster for a coffee shop. The text says Morning Blend in serif. Warm tones, natural light." That works. Adding stylistic directives and adjectives and "make it pop" language degrades the output. Where to actually use this thing: fal.ai has it at three cents per megapixel in Turbo mode. Tha

Ideogram 4.0 launched this week and within 48 hours it was available on seven platforms. That is unusual. Most model launches trickle onto one or two platforms over weeks. Ideogram went wide immediately, which suggests the open weights strategy is working as intended. Here is what you will pay depending on where you use it. fal.ai The cheapest API access. Turbo mode at three cents per megapixel. That is roughly three cents per 1K image. Balanced at six cents. Quality at ten cents. Pay-per-use, no minimums. If you are generating through an API, this is your starting point. Krea Included in all paid plans. Basic is $5.25 per month billed annually with 5,000 compute units. Pro is $21 per month with 20,000 CUs. The CU cost for Ideogram 4.0 specifically is not published yet, but Krea includes 150 plus models in their CU pool, so you are not paying extra for access. If you already use Krea for other models, Ideogram 4.0 is effectively free to try. ComfyUI Free if you have the GPU. The model is open weights at 9.3 billion parameters. Native ComfyUI support means you can download the weights and run it locally. No per-generation cost. No API calls. Just your electricity bill and GPU time. For volume generation or iteration, this is the cheapest path by far. Leonardo Announced as a day zero launch partner but the pricing page still lists Ideogram 3.0. Plans range from $12 to $60 per month with token allowances from 8,500 to 60,000. Third party models on Leonardo always consume tokens, no relaxed generation. Until they publish the 4.0 token cost, you are guessing. Assume it will be similar to their other premium models. Replicate The Ideogram 3.0 listing is live but 4.0 is not there yet. Replicate prices by hardware time rather than per-image, which can be cheaper or more expensive depending on your batch size and the GPU allocated. Worth checking when it lands. FLORA Available in FLORA. Pricing unclear. FLORA is primarily a creative platform, not an API provider, so you are

[While preparing for interviews, I realized I had a strange habit. I would solve a problem, get stuck, open the solution, understand it, and move on feeling productive. A few days later, I couldn’t solve a similar problem on my own. The issue wasn’t lack of practice. The issue was that I was consuming solutions faster than I was developing problem-solving skills. So I changed my approach. Instead of looking for answers, I started forcing myself to think longer, write down my ideas, identify where I was stuck, and only then seek guidance. That worked much better. But I couldn’t find a tool that supported this style of learning. Most platforms either: Give you the answer. Give you the editorial. Give you AI that writes the code for you. So I started building my own. The goal was simple: An AI coach that guides the thought process instead of generating the solution. Over time I added: DSA practice System Design preparation Low-Level Design preparation Company-wise interview questions Topic-wise strength and weakness analysis Personalized revision lists The interesting part wasn’t building it. The interesting part was realizing that interview preparation is less about collecting solutions and more about training how you think. What has helped you improve more during interview prep? Reading solutions? Or struggling with the problem first? Sde vault - https://sdevaultweb.onrender.com/

AI Does Not Cancel Reality I watched the conversation between Mo Gawdat and Marina Mogilko about the future of AI. The conversation is strong. It contains important ideas, but it also contains many claims that sound large in scale, although on closer inspection they rely on very broad generalizations. AI is indeed changing the labor market, education, startups, content, hiring, and ways of thinking. But it does not cancel money, connections, trust, the human vector, creativity, necessity, morality, or people’s ability to adapt. Video on YouTube AI in hiring: automation amplifies chaos Many people have entered the job market. Companies receive huge volumes of resumes. HR departments cannot handle the volume. It is natural that part of the selection process is moving to AI. But there is a serious problem here. Candidates are also starting to play against AI. Resumes are adjusted to vacancies. Cover letters are assembled around keywords. Profiles become optimized for the filter, not for real work. In such a system, the best specialist does not necessarily pass. Often, the person who understood the selection mechanism better passes. The result: the picture becomes cleaner, while the quality of the decision becomes lower. The company gets not the strongest candidate, but the candidate who matched the algorithm best. This leads to lower hiring quality, lower productivity, and slower development. “I built a startup in six weeks”: a product is not a startup The conversation includes the idea that an AI startup would once have taken years and hundreds of engineers, and now it can be built in weeks. Technically, this is true. Prototypes are now built faster. Small teams have powerful tools. One person can now do more than a group could do before. But two different things are mixed here. Building a product faster has become real. Building a startup faster has become real only when resources are present. A startup is not only code. A startup is money, connections, trust, reputa

Created a Google Chrome extension that instantly turns any highlighted text on a webpage into a Google Calendar event

Search results look simple from the outside. You type a keyword into Google, Bing, or another search engine, and you get a page of links, snippets, ads, maps, news, images, videos, and sometimes AI-generated answers. But if you have ever tried to collect search results at scale, you know it gets messy quickly. A result page is not just a list of links. It changes by country, language, device, location, query intent, and search engine. The same keyword can show different rankings in New York, London, Singapore, or Berlin. A page may include organic results, paid ads, local packs, shopping results, People Also Ask, news results, images, videos, or other SERP features. For humans, that is just a search page. For SEO teams, AI teams, data teams, and developers, it is a data source. That is where a SERP API becomes useful. What is a SERP API? SERP stands for Search Engine Results Page . A SERP API is an API that lets you collect search engine results in a structured format, usually JSON and sometimes HTML. Instead of manually searching a keyword or building a scraper to parse search result pages, you send a request to a SERP API with parameters such as: keyword search engine country language location device type output format The API then returns structured search data. A simplified response might look like this: { "query" : "best project management software" , "organic_results" : [ { "position" : 1 , "title" : "Best Project Management Software Tools" , "link" : "https://example.com" , "snippet" : "Compare features, pricing, and reviews..." } ] } This is much easier to work with than raw HTML. You can store it in a database, send it to a dashboard, compare rankings over time, feed it into an AI workflow, or generate automated reports. Why not just scrape search results yourself? You can build your own scraper. For a small test, that may be enough. You can send a request, parse the HTML, extract titles and links, and save the data. The problem starts when the workflow bec

1. update WSL environment sudo apt update && sudo apt upgrade -y 2. install dependencies If you don't use -hf option, you don't need to install libssl-dev in this step. sudo apt install build-essential cmake git libssl-dev -y If nvidia-smi shows a GPU/GPUs on your terminal, you will need to install the tooklit. This will take some time. sudo apt install nvidia-cuda-toolkit -y 3. clone the repo Build llama-cli and llama-server. This step also will take some time. If you don't plan to use -hf option, you don't need to use -DLLAMA_OPENSSL=ON . git clone https://github.com/ggerganov/llama.cpp cd llama.cpp cmake -B build -DGGML_CUDA = ON -DLLAMA_OPENSSL = ON cmake --build build --config Release # no GPU git clone https://github.com/ggerganov/llama.cpp cd llama.cpp cmake -B build cmake --build build --config Release 4. run the model Run gemma-4-12b-it with cli and server. unsloth/gemma-4-12b-it-GGUF · Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science. huggingface.co ./build/bin/llama-cli -hf unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL > hello [ Start thinking] The user said "hello" . The user is initiating a conversation. Respond politely and offer assistance. * "Hello! How can I help you today?" * "Hi there! What's on your mind?" * "Hello! Is there anything I can assist you with?" [ End thinking] Hello! How can I help you today? [ Prompt: 19.5 t/s | Generation: 11.8 t/s ] or run web-ui ./build/bin/llama-server -hf unsloth/gemma-4-12b-it-GGUF:UD-Q4_K_XL --port 8080 optional download model from huggingface mkdir -p models wget -O models/gemma-4-12b-it-UD-Q4_K_XL.gguf https://huggingface.co/unsloth/gemma-4-12b-it-GGUF/resolve/main/gemma-4-12b-it-UD-Q4_K_XL.gguf