标签：#productivity

I have 312 docs in my personal knowledge DB. Tweets, arxiv abstracts, Zenn articles, blog posts, YouTube transcripts. Claude Code writes to it, reads from it, and cites out of it every day. That number is not a brag. It is the reason I finally have data on which retrieval strategy holds up in an LLM-native workflow. I tried four. The one I ship is the one I tried last and expected to lose. Three of the four broke in ways that are worth naming, because the broken versions are what most tutorials will tell you to build. The setup, so we agree on what got benchmarked The knowledge DB is called context-forge internally. It is a folder, some markdown files, and a SQLite table. Claude Code adds to it via CLI, searches via CLI, and reads the underlying markdown directly when it needs the full text. It took eight hours to build the CLI, three months to accumulate the 312 documents at a pace of one to five per day, and about 15 minutes a day of my time to keep it flowing. Each doc has metadata: source URL, a credibility score 1-5, one to three categories, a short summary. The autoregistration pipeline is Claude Code itself: I paste a URL, it fetches, summarizes, scores, categorizes, writes the markdown, commits, and updates the SQLite index. The pipeline is not the interesting part. The retrieval strategy is. I ran each of the four strategies for two weeks against the same day-to-day tasks: writing a chapter, answering "what did that person say about X," and building an argument for a decision. Same me, same DB, different retriever. Layer 1: pure semantic RAG (vector embeddings). Broke at 200 docs The first version was the textbook answer. Embed every document with a sentence transformer, store the vectors in SQLite with a similarity index, retrieve the top-k on every query. This is the pattern Silicon Slopes covers for code-level RAG and Anthropic itself has an issue open for a built-in version . It worked at 50 docs. It worked at 100. Around 200 documents it started retrie

Every software agency has been here: the client asks for a price, you give a range (say $45k–$65k), and two things can happen. Either the client nods and you win the deal at the low end — or they get suspicious and ask "so you don't actually know how much it costs?" Range pricing is often misunderstood. Used wrong, it looks like you're guessing. Used right, it's the most honest and professional way to price software projects — because anyone who gives you a single fixed number for an undefined project is either padding heavily or gambling with their margin. This guide covers when to use range pricing, how to structure it, and — most importantly — how to present it so clients trust you more, not less. Why Single-Point Pricing Is a Problem A fixed price for an undefined project forces you into one of two positions: You pad aggressively — add 40% contingency, quote $70k for a project you'd happily do for $50k. If the scope doesn't expand, the client overpays. If it does, you're protected. Either way, one party loses. You guess lean — quote $50k based on your best assumptions. If the client adds features mid-project, your margin evaporates. The client thinks they're paying for X, you're building X+Y. Both parties end up frustrated. A pricing range avoids both traps. It says: "based on what we know today, this project falls between $45k and $65k. Here's what needs to be true for the low end, and here's what would push it toward the high end." That's not guesswork. That's transparency. The Anatomy of a Good Pricing Range Not all ranges are created equal. A useful range has three properties: 1. Width That Respects Uncertainty The width of your range communicates how well you understand the project. Range width What it signals When it's appropriate < 15% ($50k–$57k) High confidence Detailed spec, similar past projects, known team 15–30% ($50k–$65k) Moderate confidence Clear brief, some unknowns in tech or integration 30–50% ($50k–$75k) Low confidence Vague brief, new domain

A practical guide to writing specs that turn vague AI prompts into production-quality code — from functional requirements to edge cases, with real before-and-after examples. Let me paint a picture you've probably lived. You open Claude Code or OpenCode, type a vague prompt like "add a user dashboard with analytics," and hit enter. The agent spins up, writes a bunch of code — it even looks decent at first glance. Then you realize: the data model is wrong, the API endpoints don't match your existing patterns, there's no error handling, and the "analytics" is just a row of four hardcoded numbers. You spend the next hour correcting, prompting, correcting again. You would have been faster writing it yourself. Now imagine a different scenario. You spend 15 minutes writing a structured specification, paste it into the agent, and it produces exactly what you wanted — following your conventions, handling edge cases, wired into your existing auth and data layer. One shot. That's not luck. That's the difference between treating your AI coding agent like a chatbot and treating it like a senior engineer who needs a clear design document — also known as spec-driven development . The Core Problem: AI Agents Are Powerful but Undirected Claude Code, OpenCode, Cursor Agent — these tools are incredible when pointed at a well-defined task. They can read your entire codebase, understand your conventions, and produce production-quality code. But they share a fundamental limitation: they don't know what you want unless you tell them, precisely and completely. When you give an agent a one-liner prompt, you're leaving an enormous amount of ambiguity. The model will fill in the gaps — but it fills them with its own assumptions, which are often generic, incomplete, or just wrong for your context. A spec closes those gaps. It transforms an open-ended creative writing exercise into a constrained engineering task. What Makes a Spec Actually Work for AI Coding Agents? A spec that works for an AI

The most valuable thing Claude has done for my work is help me make better architectural decisions before I start building. When I was designing my hotel reservation system I needed to handle multi-currency pricing and timezone conversion. My first instinct was to put all of that logic directly in the booking controller alongside the reservation code. It would have worked but it would have created a mess that was hard to test and harder to extend. Before writing anything I described the problem to Claude and talked through a few approaches. The conversation helped me see that separating pricing and timezone logic into dedicated service classes would make each piece independently testable and easier to swap out later. The booking controller would just call those services without knowing how they worked internally. That conversation took maybe twenty minutes. It saved me hours of refactoring later. I still designed the system. I still made the judgment calls. But the quality of my thinking going into implementation was significantly better because I had a thinking partner to pressure test my ideas against.

The problem In 2026, devs spend 11.4 hours a week reviewing AI-generated code — more time than they spend writing it. We're burning cycles fixing bugs our own AI tools wrote. I started calling this "AI debt": the maintainability tax that piles up when nobody's actually reading the code the assistant just spat out. I wanted a fast, brutal way to see how much debt was hiding in a file before I even opened a PR. What I built Roast My Code — paste a code snippet, get an AI Debt Score (0–100) and get roasted for your sins. 118 regex patterns across 8 languages (JS/TS, Python, Go, Rust, Java, PHP, C++) Scores broken into Readability, Structure, Error Handling, Safety, and Style Code metrics: nesting depth, duplication %, comment ratio, avg line length Three brutal one-liner roasts + concrete fixes for each issue found The twist: zero AI. No API calls, no LLM, no backend. Everything runs client-side with regex pattern matching. Your code never leaves your browser. Why regex, not AI Honestly — irony. A tool built to call out AI slop shouldn't itself be another wrapper around GPT. Regex is also just... faster. No API latency, no cost, no rate limits, no "please wait while I analyze your code" spinner. You paste, you get roasted in under a second. It's not going to catch everything a proper linter or an LLM code reviewer would. That's not the point — it's a gut-check, not a static analysis suite. A taste of the roasts javascript var API_KEY = "sk_live_51H8xJ2kL9mNpQrStUvWxYz..."; if (a == 1) { if (b == 2) { if (c == 3) { x = eval(a + b + c); } } } 🔒 is that a hardcoded credential? in 2026? we need to talk. your teammate rewrote this on a Sunday. FIX: Move it to an environment variable or secret store, then rotate the credential. 🎆 eval(). we don't need to say more. you know what you did. this is the part reviewers skim past. FIX: Replace eval with a lookup table, JSON.parse, or an explicit parser. Try it 🔗 Live app 💻 Source on GitHub — MIT licensed, PRs welcome Paste your wor

A few days ago, I wrote "Understanding Over Origin" and it got alot of engagement and I'm really...

A few months back I was running Sales Navigator searches for a client project — filtering down to "VP Sales, fintech, based in Italy or Spain" type lists — and the results were genuinely good. 60, 80 leads that actually matched. Then I hit the part nobody warns you about: there's no button on that page that says "save this." So I did what everyone does. Opened a spreadsheet, alt-tabbed back and forth, typed names and job titles by hand. Around profile 40 I gave up and went looking for a better way. This is what I found, roughly in the order I found it, including the tool I ended up building because none of the existing options quite fit what I needed. First: the export LinkedIn actually gives you LinkedIn has a real, built-in data export, and most people don't realize how narrow it is. It's under your profile photo → Settings & Privacy → Data Privacy → Get a copy of your data . From there you either tick specific categories (that email usually lands within minutes) or request the full archive, which takes closer to a day and sometimes arrives in two batches. Either way you get a download link that expires after 72 hours — and it's desktop only, the mobile app won't let you request one. What you get back is genuinely thorough: connections, messages, your own profile history, activity, even the ad-targeting data LinkedIn holds on you. A couple of quirks worth knowing before you rely on it: some connections' email addresses will just be missing, because sharing an email on download is something each person opts into individually, and you won't get a list of who viewed your profile or any "People You May Know" data. If you're in the EU, EEA, or Switzerland, LinkedIn also runs a separate API for pulling your data on a schedule rather than as a one-off request. Here's what this export is not built for, though: it has no idea what you searched for yesterday. It's an archive of your own account, not a way to capture a live search. Run a Sales Navigator query and pull 80 lea

Listmargin works out your eBay final value fees, ad fees, and what you actually keep on a sale. Most fee calculators copy a four-line summary of eBay's rates. This one reads the full published schedule: all 46 category rates, the store subscription tables, and the four categories where crossing a price threshold re-rates the entire sale. It's free, with no signup and no paid tier. There's an embeddable version if you run a blog or a tool site. A weekly monitor re-reads eBay's own fee pages, so when a rate moves the calculator gets corrected instead of drifting out of date.

Two numbers from my own systems, side by side. When I deploy, a check confirms the pages are actually live. It retries three times, twenty seconds apart, so it tolerates up to a minute of "not there yet" before calling anything wrong. When my monitor decides whether production is down , it waits five seconds and retries once. The check that guards the more consequential claim is the more impatient one. I did not decide that. I never compared them. Until last week I had never seen those two numbers in the same place, and neither had anything else. Where the numbers came from The deploy one has an origin story I'd have told you proudly a week ago. I shipped nine pages, then checked the URLs immediately instead of trusting the CLI's success message. Four returned 404. Nothing was broken — CDN propagation — and twenty seconds later all nine were 200. A single check at the wrong moment would have told me, with total confidence, that a perfectly good deploy was broken. So I wrapped it in a retry loop. Three attempts, twenty seconds apart. Problem solved, and it even sounds like engineering. Here is the part that isn't. I picked twenty because it was the first interval where the false alarms stopped. My sample was about three deploys. I have never recorded how long propagation actually takes. I widened the tolerance until the red went away, and then I wrote about it as if I'd learned something. Someone in a thread named this before I saw it: a tolerance chosen that way is the same muting I'd been criticizing, relocated inside the assertion where it reads as rigor instead of avoidance. The test I was given, and the answer I didn't want In that same thread I speculated that my deploy tolerance was probably leaking into my outage detector through a shared helper. It sounded plausible and I said it like a finding. The reply was sharper than the guess: that's a falsifier, not evidence. Here's the concrete test — do the two checks consume the same retry policy or threshold confi

You know exactly what you want. You can see it. You just don't know what it's called. So you open your AI tool and type "pixelated fade effect". Then "retro dot gradient". Then "that grainy old-computer image style". Six rounds later you have something almost right, and almost right is the most expensive kind of wrong. The word was dithering . With it, one prompt gets you the real thing. This gap has a shape. AI collapsed the cost of building, so the bottleneck moved: it's no longer "can the AI do it", it's "can you name it". Every field you touch as a builder has a precise vocabulary, and the words you're missing are costing you rounds of generation, wrong libraries, and vague briefs. Vocabulary is the highest-leverage thing you can pick up right now, and nobody teaches it. Jargon Buster is the cure. It's a free reverse-lookup glossary built for exactly this moment: you describe the thing in your head, it gives you the word. Reverse lookup: describe it, get the word Press Cmd+K on any page and type what you'd say to a colleague, not the term: What you type What you get "the glowy circles behind them" Bokeh "the grid of differently sized cards" Bento grid "the scroll that takes over the page" Scrolljacking "grainy speckles when I turn the number up" ISO "why is my payout smaller than my sales" Settlement "the inside of the letters fills in when I bold it" Counter Misspellings work too. "Ditter" lands on Dithering. That's deliberate: the fuzzy phrasings and typos people actually reach for are stored on every entry as first-class search data, not errors to correct. A normal glossary is indexed by the words you don't know. This one is indexed by the words you do. Every entry ends prompt-ready Knowing the term is half the loop. Each of the 2,142 entries closes the other half: A plain-language one-liner for the "that's the word!" moment A short explainer : what it is, when to reach for it, the gotcha A prompt-ready snippet : the concept translated into an instruction an

Handify ai Like many side projects, this one started because I had a simple problem to solve. I wanted a way to convert digital text into realistic handwritten notes without spending hours writing everything manually. Most existing tools I tried either looked too robotic or offered very little customization. So I decided to build my own. The Goal Instead of just changing a font, I wanted the output to actually feel handwritten. Some of the features I focused on were: 📝 Convert typed text into realistic handwriting 📄 Upload your own notebook or paper template ✍️ Multiple handwriting styles 🔀 Mix two handwriting fonts for a more natural appearance 🎲 Character variation so repeated letters don't always look identical 📥 Export high-quality PDFs ready for printing Challenges Making handwriting look "real" is much harder than simply rendering a handwriting font. Some of the biggest challenges were: Preventing repeated letters from looking identical. Keeping line spacing and word wrapping natural. Supporting different paper templates. Generating high-resolution PDFs without losing quality. Making the experience fast enough to generate pages within seconds. Small details make a surprisingly big difference when people compare AI-generated handwriting with actual handwriting. * Tech Stack * The project is built using: React TypeScript Firebase Vite Capacitor (Android App) Google Analytics What I Learned Building the product was only half the work. The harder challenge has been: SEO Google Search indexing Play Store optimization Improving conversion rates Understanding user behavior through analytics A great product doesn't automatically get users—you also need to make it discoverable. Current Progress The project is still growing, but it's already receiving organic traffic from Google and users have started using it for: Study notes College assignments Personal journals Printable handwritten documents Seeing people use something you built is incredibly motivating. I'd Love Yo

A few weeks ago Boris Cherny, who leads development on Claude Code, mentioned during a talk at Acquired Unplugged that he doesn't really write prompts for Claude anymore. Instead he writes loops that keep prompting Claude until the work is actually done. The clip went viral on X, racked up nearly 700k views in under 24 hours, and Loop Engineering became the latest term making the rounds in AI development circles. The core idea is straightforward enough. Rather than obsessively tuning a single prompt to get a perfect output on the first try, you build an iterative system around the model: give it a clear goal, feed it the right context, give it tools to work with, evaluate what it produces, and define conditions for when it can stop. Wire those pieces together and the agent stops being a one-shot call and becomes something that iterates, self-corrects, and keeps working until the output actually meets your bar. The efficiency gains over prompt-tuning are real, and that is why the concept resonated so quickly. What struck us as we built and shipped the loop system for our own platform Octo is that almost all of the current conversation around Loop Engineering stays at the single-agent level. You have one model, one cleverly designed loop, one sandbox, and the agent grinds away iteratively until its output passes whatever checks you have set up. That solves a real problem: how one person works faster with AI. But real work, especially inside an organization, rarely fits cleanly inside a single agent loop. A product feature going from idea to shipped code needs someone defining requirements, someone designing the approach, someone writing the implementation, someone verifying quality, someone feeding back results. Those are not different iterations of the same loop. They are interconnected loops that need to pass context and outputs between each other. When loops need to share state, trigger each other, and respect organizational boundaries, single-agent loop design sto

Building Three Privacy-First Mini Apps That Feel Like Standalone Products PureHub is an open-source collection of 22 free, ad-free mini apps. This release focuses on a simple product question: can a mini app inside a hub still feel dependable, focused, and complete? QR Studio The web scanner now supports a live camera and uploaded images through local decoding. Scan history stays in local storage, URL results receive basic safety checks, and supported cameras expose a torch control. Android uses CameraX and ML Kit with explicit scanner cleanup, duplicate-result protection, and copy, open, and share actions. Zen Pomodoro A one-second decrement loop drifts when a tab sleeps. The new timer stores a target time and recalculates the remaining duration, so switching tabs or waking a device no longer quietly extends a session. Weekly sessions and focused minutes remain on-device. Android uses a monotonic clock for the same reason. Zen Breath The breathing guide now includes Calm 4-6, Box 4-4-4-4, and Relax 4-7-8 patterns, controlled sessions, cycle totals, and accessible motion behavior. Nothing requires an account. Standalone safety for all 22 tools Each mini app now has a runtime contract describing its local storage namespace, offline behavior, and device capabilities. A per-tool error boundary prevents one failure from taking down the rest of PureHub. The three flagship tools also load as independent chunks and are available as PWA and Android launcher shortcuts. What happens next The Command Center will compare 14 days of anonymous aggregate opens, helpful votes, and shares. The strongest useful-use signal - not raw views - will choose the next deep-polish target. Try the release at PureHub or inspect the source on GitHub .

How one frustrated shopper turned a spreadsheet nightmare into a community-powered resource that's saving buyers hours of research. If you've ever shopped for a mini PC, you know the drill. Open a dozen retailer tabs. Copy specs into a spreadsheet. Squint at product titles that hide critical details. Try to figure out whether that $299 model has soldered RAM or upgradeable slots. Give up and buy something you're not quite sure about. This is exactly the frustration that led a developer to build Mini PC Price — a free, sortable comparison table that aggregates real marketplace listings with full specifications, all in one view. The Problem: Specs Buried in Fine Print The mini PC market has exploded in recent years. Brands like Beelink, Minisforum, Intel NUC, ASUS, Lenovo, and HP are flooding the market with devices ranging from $100 stick-class boxes to $1,000+ workstations. But the shopping experience hasn't kept pace. Most e-commerce listings show a product name, a price tag, and a few bullet points. Critical information — RAM type (soldered vs upgradeable), GPU class (integrated vs discrete), OS bundle (Windows pre-installed or barebone), storage interface (NVMe vs SATA) — is often buried in product descriptions or missing entirely. "The difference between an 8 GB listing and a 16 GB listing of the same model can be easily missed if the table doesn't make it obvious," the creator explains. "I wanted to build a tool that surfaces exactly what matters, side by side, so buyers don't have to guess." The Solution: One Table, Full Specs, Community-Maintained Mini PC Price is not another curated top-ten list or a simple deal feed. It's a full-parameter database where each row represents a real marketplace listing. Prices refresh daily, and specs are collected from official product pages. Critical attributes receive human verification where mismatches are common. The tool's default view sorts by price ascending — the fastest way to find mini PCs under $100 or $150. But th

First hour with Claude's browser extension: I pointed it at our LLC registration and watched it work through the forms, finding the right pages, filling the fields, moving on. I sat there holding a coffee, contributing nothing. I thought: I'm going to ship so many products. I shipped one. Here's what happened in between. Everything below was caught before launch. None of it was caught by being clever. It was caught by a process that got built slowly, mostly after being burned. What went wrong The idea wasn't the hard part. Once I went looking, I found several products with some of the same features. Nobody had the exact combination, but the idea was never the moat. Good implementation and distribution seem to be. You design the product while building it. Referral behaviour, what happens when a trial expires mid-session, how translations work across a page, none of it was in my head at the start. Each became a decision made under pressure, halfway through something else. Write as much of the workflow down as possible first. It says it did things it didn't do. Confidently. I deployed more than once to find the fix I'd been told about was never written. Treat every claim of completion as unverified. The rule that came out of it: make it prove the code is right before it theorises about what you did wrong. Bullet points, not paragraphs. Long replies made it hard to tell which of my five points got addressed. Numbering my instructions, and making it map answers back to the same numbers, turned "did you do item 3" into a question with an answer. It blames you first, and argues with facts. Two landing page changes; one appeared, one didn't. Its verdict: "you didn't deploy." I said one change was live, which is only possible if I had deployed. It repeated that I hadn't. It never asked which change I could see, and never reopened its own code, where the bug was. I swore at it. It stopped guessing, checked, and found the error. Many times, escalation seemed to be the only thi

I remember the day I hit my breaking point. I had spent the entire morning — five hours — wrestling with a React component that an AI assistant had generated for me in about four minutes. The code looked flawless at first glance. Proper hooks, clean JSX, even decent comments. But it didn't work. And worse, I couldn't figure out why. Everyone talks about how AI speeds up coding. And it's true — when it works, it's magical. I've personally seen my feature delivery time drop by maybe 40-50% on good days. But what nobody talks about — what I certainly never saw in the breathless LinkedIn posts — is the debugging nightmare that follows when the AI gets it wrong. That day, I realised I had spent ten times longer debugging AI-written code than I would have spent writing it myself from scratch. I started tracking it. Over three months, I logged every AI-assisted task. The numbers were sobering: on average, each AI-generated snippet took me 3.2 times longer to verify and fix than to write myself. And for complex tasks — anything involving state management, async flows, or edge cases — the ratio jumped to 8-12x. The AI was giving me confidence, not correctness. And confidence, as any seasoned developer knows, is the enemy of debugging. The Hallucination That Cost Me a Sprint One incident stands out. I was building a data pipeline in Python that needed to batch-process JSON files from an S3 bucket and push transformed records into a PostgreSQL database. I gave the AI a detailed prompt: "Write a function that reads all JSON files from a given prefix, validates each record against a schema, and inserts them in batches of 500. Use threading for I/O." The AI returned a beautiful 60-line function. It used concurrent.futures.ThreadPoolExecutor , had proper error handling, even logged progress. I was impressed. I dropped it into the codebase, ran the tests — they passed. Deployed to staging. Worked like a charm. Then production hit. Three hours later, the database had 30,000 duplicat

I built a free ATS resume checker after seeing too many friends get ghosted on Easy Apply. Most people think ATS means “add more keywords.” In practice, a lot of resumes fail earlier — on parsing. The 5 blockers I see most Columns / two-column layouts Nice for humans. Bad for many parsers. Contact info or skills in a sidebar often get read in the wrong order (or skipped). Fix: one-column layout. Standard section titles: Experience, Education, Skills. Icons instead of text Phone / email / LinkedIn as icons only = empty fields for the ATS. Fix: real text next to icons (or drop the icons). Text in images / fancy graphics If you can’t Ctrl+A → copy readable text from your PDF, neither can the ATS. Fix: real text PDF/DOCX, not a designed image export. Tables for experience Tables often scramble dates and bullets. Fix: simple headings + bullet points. Keyword stuffing Repeating “Python Python Python” can look spammy and still miss role-specific terms from the job post. Fix: mirror the job’s real skills in your bullets with proof (tools + outcomes). Quick self-test Open your PDF Select all → paste into a blank doc If the order is messy or text is missing, fix the format before you apply again Free checker I use for this I shipped a free diagnosis that flags these blockers first (full score/keywords optional): https://myatscheck.com/free-ats-resume-checker If you’re job hunting through Workday / Greenhouse / Easy Apply, fix parsing before you rewrite every bullet.

This was first published on the Vibsync blog . Reposting for the DEV community. The short version: AI reliably makes each developer faster. Whether it makes the team faster is a separate question — and the gap between the two is where a lot of quiet cost hides. Below: the five coordination costs that eat the difference, a ten-question diagnostic, and five operating principles. Picture three developers, three AI coding agents, and one repository. Each developer can now produce candidate code, tests, and refactors faster than before. Yet releases move at the same pace, review queues grow, and the same facts keep getting rediscovered. That's not a paradox, and it isn't a reason to slow anyone down. It's a reminder that individual speed and team speed are different quantities , and AI coding agents scale the first far more easily than the second. Give everyone a faster typewriter and you get more pages — not necessarily a better book, written faster, by a group. Individual output is not team throughput It's worth separating two things we tend to blur: Individual output — how much finished work one developer (plus their agent) produces. Team throughput — how much shippable, coherent work the group produces together, after review, rework, waiting, and reconciling everyone's changes. AI agents lift individual output directly. Team throughput is what's left after the coordination overhead is paid, and that overhead doesn't shrink just because each person got faster. A useful way to hold it in your head — not as a formula to compute, just as a shape: team throughput ≈ the sum of local speed-ups − rework − waiting − reconciliation When you add agents, the first term grows. If nothing else changes, the last three grow too — because there's now more work in flight, produced faster, by people who can't all see what the others are doing. The interesting question for a team lead isn't "how do I make everyone faster?" It's "which of those subtraction terms is my real ceiling?" Ther

A few months ago I got frustrated. I needed to compress a PDF quickly. Found a tool online — it asked me to create an account first. Found another — it had so many ads the actual button was invisible. Found a third — it uploaded my file to their servers and I had no idea what happened to it after. I thought: this shouldn't be this hard. So I built EazyStudio — a suite of 80+ browser-based tools where everything runs 100% in your browser, no signup, no intrusive ads, no files ever leaving your device. What's inside Here's a snapshot of what's available: PDF tools Compress, merge, split, rotate PDFs PDF to Word, Excel, JPG and back Add watermarks, protect with passwords Image tools Background remover (runs locally in browser) Image compressor, resizer, converter AI image upscaler Color palette extractor, color picker Developer utilities JSON formatter/validator Base64 encode/decode URL encoder, HTML entity converter Regex tester JWT decoder API tester (Postman-lite) CSS gradient generator, box shadow generator Finance & math EMI calculator, SIP calculator GST calculator, compound interest, tip splitter Unit converters (length, weight, temperature, data) And more QR code generator Password generator Text tools (word counter, case converter, lorem ipsum) Device preview tool The technical approach: browser-first The biggest design decision was: nothing gets uploaded to a server. For PDF operations I use PDF.js and pdf-lib running in the browser. For image tools it's canvas + WebAssembly (WASM) modules. For background removal I'm using a WASM-based segmentation model that loads client-side. This has three benefits: Speed — no upload round-trip, works on large files instantly Privacy — your files never touch my server Cost — zero storage, zero egress bandwidth The downside: WASM modules add initial load time. I worked around this with lazy-loading — the WASM only loads when you first use that specific tool. What I learned building this 1. People hate signups more than I ex

Hello Everyone, I'm the creator of PyBotchi, an intent-based AI Agent Orchestrator. In this post, I will discuss some key concepts why I created it. A little bit of background first. I'm a solutions architect with 10 years of experience as a software engineer. Most of my work are high throughput, high reliability, low cost and low latency services. This is while making it simple and readable to improve it's maintainabality. When I'm designing a system, I usually prioritize these concerns. You may assume this is my bias in relates to AI Agent building. I'm also Claude Certified Architect (Foundation) and I found that PyBotchi aligns almost identical to Anthropic's core agent recommendations. TL;DR: PyBotchi is an lightweight, async-first Python framework that uses nested Pydantic models and OOP inheritance to turn LLM intent detection into clean, deterministic business logic without the overhead of complex graph orchestration. Why I created PyBotchi? I really believed that traditional coding can already solved what client's need. The only limitations we have is how we read the input and how we show the output. In most cases in web services, your API use JSON, XML, etc with their respective specification/structure. Input Analogy Assume you have created a Books CRUD endpoints (FastAPI with Pydantic). Your create endpoint will have a define specifications for book creation to have a validation and avoid user errors. Most of the time you will also validates sessions and permissions which also included in the request. If you want your chat bot to support those, you just need add those endpoint as intent (tools). If your model tool selection are able to detect intents. You are more "close" to being deterministic. "Your services will have 50 endpoints or more. You will flood your tool selection call" In your frontend UI, you segregate panels/forms/inputs in their respective pages. You don't usually join multiple intent in a same page. Cluttered UI will make your UX confusin