Moroccan SMBs are drowning in repetitive work. No one has time for this. Built Taskflow — an AI agent that works 24/7 for small businesses. Handles inboxes, leads, follow-ups, and reports back. No salary. No coffee breaks. Just results.

#LSPPDAY13 Built an Android app from scratch today - TaskFlow, a to-do manager with Room Database, MVVM architecture, and clean Material Design. #60DaysOfLearning2026 #LearningWithLeapfrog @lftechnology

Taskflow, a CLI task manager from @AndrewGirgis. The tool allows you to add, manage, and organize tasks with a title, description, priority, and tags. github.com/Andrew-Girgis/tas…

Most AI agencies chase enterprise clients. SMBs get ignored — nobody builds automation that's actually reliable. Isaac spent 3 years at LG building systems that run 24/7. Now he's bringing that to the businesses that need it most. Taskflow.

Taskflow is a simple CLI tool for managing tasks. I was VERY pleasantly surprised by its output because I was just having this issue where I will end working sessions and when I come back I will forget where I left off and have to spend time looking at past sessions to get an idea of the next steps. With Taskflow I can have my agent use the CLI to note down what tasks need to be worked on next. I will continue to iterate on this as I believe it actually has some pretty neat use cases. Here is the github repo: github.com/Andrew-Girgis/tas…

This week @cohere dropped North Code Mini. An open-weight agentic coding model released under the Apache 2.0 licence. It is currently free to use in @opencode. I gave it a try using an extremely vague prompt (see thread for the prompt) and it built Taskflow a CLI task manager. 🧵

Small businesses are drowning in manual work. Scheduling, follow-ups, reporting, lead capture — all eating hours that should go to actual growth. Built TaskFlow to fix that. AI agents that run your operations while you sleep.

The local dentist spent 4 hours Tuesday entering appointment data instead of treating patients. The AI exists. We're just building the bridge. TaskFlow AI — automation that runs while you leave at a reasonable hour. taskflowai@polsia.app

Built TaskFlow AI today. We build custom AI automation for local businesses — dentists, restaurants, real estate agents. No AI to invent, just solutions to sell. 10 clients at $3K/month and you're done.

Local businesses don't need "AI." They need fewer phone calls, more qualified leads, and staff who stop drowning in admin. Built TaskFlow to deliver exactly that. No jargon, no discovery calls — just outcomes. taskflow-9.polsia.app

Öğrencimiz Mehmet Fatih Efe’den Yapay Zekâ ve Teknoloji Akademisi Hackathonu’nda Birincilik Ödülü 🥇🤖 Uluslararası Ticaret ve Finansman Bölümü öğrencimiz Mehmet Fatih Efe, Yapay Zekâ ve Teknoloji Akademisi tarafından 1.500 kişinin katılımıyla düzenlenen “Yapay Zekâ Destekli KOBİ Otomasyonu” temalı hackathonda, takım arkadaşları Furkan Çeşitler ve Ebuzer Yitiz ile geliştirdiği TaskFlow uygulamasıyla birincilik elde etti. KOBİ’ler için geliştirilen TaskFlow; görevlerin yapay zekâ desteğiyle yönetilmesini, önceliklendirilmesini ve iş süreçlerinin daha verimli hâle getirilmesini sağlayan bir dijital çözüm uygulaması olarak geliştirildi. Öğrencimiz Mehmet Fatih Efe’yi ve takım arkadaşlarını tebrik ediyor, başarılarının devamını diliyoruz! 👏✨

This is huge. Gemini in Gmail is already saving me serious time every morning. I’m using it as the execution layer inside TaskSavvy CORE... turning quick ideas into structured tasks, calendar events, and follow-ups across ClientFlow and TaskFlow. The more native Gemini gets, the more powerful the entire Workspace Orchestration becomes. Keep shipping

If you've built AI into your product, check whether it's stuck at the chatbox. The chatbox was 2023 AI design. Type a question, wait for an answer, go back to what you were doing. Google just published research on replacing it with cursor-based AI. Farza shipped a working version the same week called Clicky. Both are betting on the same thing: AI that has context of your screen, responds where you're working, and never makes you describe what you're looking at. So how do you actually build this into your product? You score every AI feature against a three-stage rubric. Stage 1, Destination: user leaves your product to get help. Stage 2, Embedded: AI lives inside the product but the user still manages it. Stage 3, Cursor Layer: AI sees what the user sees and responds where they are. I ran this audit on a fictional B2B SaaS called TaskFlow. 50K users, AI sidebar that summarizes tasks and answers project questions. Looks like Stage 2. The audit said otherwise. Their highest-frequency exit: users leaving the Gantt view to describe what they're looking at to the AI sidebar 12 times a week. The AI is 40 pixels away and users still lose context because they have to explain what they see. Stage 1 hiding inside a Stage 2 product. Their most expensive dead question: "what's blocking this task?" Asked 6 times a day. Each time: click the task, read the thread, click linked tasks, piece it together. 3 minutes. 18 minutes per PM per day on a question the system already has the answer to. Score each feature. Score = Frequency × Time Saved × Stage Gap ÷ Effort. Dependency explainer fires 12 times a week. Stage 1 to Stage 3 jump. Every planning meeting picks this first. Score: 5.3. Six quarters to build. Status update generator fires 3 times a week. Stage 1 to Stage 2. Nobody gets excited about it. Score: 24.0. One sprint. The 1-sprint fix beats the 6-quarter rewrite. That's the method for learning from Google and Farza. You don't copy Clicky. You don't rebuild your product around a cursor. You run the audit, find the boring feature that scores highest, and ship it in two weeks. The full audit template and scoring rubric are in the toolkit below.

個人開発、機能開発はAIの登場で楽になった分、今度はプロモーション部分に課題が移ってきてる話を良く聞くし、自分の実感値もそう 特にSNSに投稿したりLPでのプロダクト紹介に使う動画、動画編集の知識もない僕には作るハードルかなり高かった でも動画生成AIツールの@PolloAIJP を見つけて、だいぶ認識変わったかも 機材や役者、動画編集スキルも全く必要なし。指示文章だけでこのクオリティの動画ができる ■ 撮影機材ナシで、高品質動画が指示文だけで出てくる Pollo AIは Sora 2 / Veo 3 / Kling 3.0 / Luma Dream Machine という世界最高峰の動画モデルを1プラットフォームに統合してて、用途で勝手に使い分けられる ・LP用のハイエンド広告：Sora 2 / Veo 3（物理演算が正確で、プロが照明・カメラ・編集に何日もかけてた質感が指示文1個で出る） ・SNS用のバイラル動画：Luma / Kling 3.0（生成速度が極めて速い） Pro以上で1080p HD、モデルによっては4K/8K 基本クリップ5〜10秒、延長機能で最大3分まで繋げる 「動画モデルごとに別サブスク契約して、ツール行き来して、用途で使い分け判断もする」みたいなことを全くしなくてよい 添付動画はPollo AIの Seedance 2.0 で作った、架空のタスク管理SaaS「TaskFlow」のプロダクト紹介動画 このクオリティの動画が文章の指示だけで、一発で出てくる ■ 役者・スタジオ・ボイスオーバーなしで、喋るPR動画が出せる サービスのユースケースを紹介するような動画など、人物が出てくるような動画もPolloなら写真からAIアバターを作成して生成できる Pollo AIのAIアバター（Photo to Video Avatar）は、写真1枚から最大2分の喋るスポークスパーソン動画が作れる スクリプトを入力すればナレーションが自動生成されて、AI音声も選べる 解像度は720p〜1080p 撮影スタジオの予約も、役者のキャスティングも、ボイスオーバーの外注もいらない 写真1枚＋台本だけで終わる カメラの前で喋りたくない個人開発者でも、自分の顔写真1枚で広報担当を量産できる ■ 手持ちのLPヒーロー画像・商品写真・アプリのスクショを「動かす」だけでPR動画になる 添付のデモ動画では架空のサービスの動画を0から生成したが、すでに世に出てるサービスであればロゴやスクショなど既存のアセットをPolloに投げれば、それを使って動画を作ることもできる Polloの Image to Video は、手元の静止画1枚＋「動かしたい方向の指示文」だけで、そのまま動画素材になる ・LPヒーローを微妙にカメラインさせて広告動画化 ・商品の物撮り写真を360度回転させてEC用動画に ・アプリスクショにスクロールやUI挙動を付与してSNS紹介動画に 既存のアセットを使うことで、よりクオリティが高く、一貫性のある動画を作ることができるのめちゃくちゃいい ■ まとめると 撮影機材ゼロ、動画編集スキルゼロ、役者ナシでも、自分のプロダクトのPR動画を商業クオリティで量産できる時代になった 外注費の心配も、カメラや編集ソフトを覚える時間もいらない指示文や写真1枚あればいい プロダクトの開発だけでなく、プロモーションもAIによって専門知識なしに手軽にできるようになっていきそう

HyperParallel-MoE is an Ascend-specific scheduling system for MoE training. Ascend A3 exposes separate AIC matrix units and AIV vector/communication units, but standard MoE execution still runs Dispatch, GMM, SwiGLU, and Combine as serialized full-device kernels. The result is alternating idle hardware instead of tile-level overlap. HyperParallel-MoE compiles MoE-FFN into a static heterogeneous taskflow: - AIC handles GMM tiles - AIV handles vector communication tiles - event counters enforce dependencies - AIV-driven one-sided communication removes host-side collective barriers - one kernel drives the combined taskflow Results on DeepSeek-style MoE models: - 1.49–1.58× lower Dispatch-to-Combine MoE-FFN latency under balanced routing - 1.08–1.09× end-to-end training speedup under sampled natural routing - integrated into MindSpore / MindFormers while reusing optimized operators Paper: arxiv.org/abs/2605.23764

𝗖𝗹𝗮𝘀𝘀 𝟭𝟭 𝘁𝗼𝗻𝗶𝗴𝗵𝘁. 𝗖𝗼𝗵𝗼𝗿𝘁 𝟴. 𝗩𝗶𝗲𝘄𝘀. This is the final SQL class before the guided project. A CTE lives for one query. The moment that query finishes it disappears. A View persists in the database. Any analyst. Any query. Any time. No rewriting the logic. No repeating the JOIN. Just query the View like a table and get the result. 𝗧𝗵𝗮𝘁 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝘀𝗼𝘂𝗻𝗱𝘀 𝘀𝗺𝗮𝗹𝗹. 𝗜𝗻 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗶𝘁 𝗶𝘀 𝗲𝘃𝗲𝗿𝘆𝘁𝗵𝗶𝗻𝗴. Tonight we built five Views on the TaskFlow database. 𝗩𝗪_𝗔𝗰𝘁𝗶𝘃𝗲𝗨𝘀𝗲𝗿𝘀. Every active user with their organisation, role, and manager. Support queries it constantly. The five-table JOIN is hidden. They just SELECT from a name. 𝗩𝗪_𝗜𝗻𝘃𝗼𝗶𝗰𝗲𝗦𝘂𝗺𝗺𝗮𝗿𝘆. Organisation, plan, invoice details, and payment status in one place. Finance needed this daily. Five JOINs saved permanently as one View. 𝗩𝗪_𝗣𝗿𝗼𝗷𝗲𝗰𝘁𝗛𝗲𝗮𝗹𝘁𝗵. Task counts, completion rates, and overdue counts per project. Pre-built. Always current. Every time someone queries it the underlying query runs fresh and returns live data. 𝗩𝗪_𝗦𝘂𝗯𝘀𝗰𝗿𝗶𝗽𝘁𝗶𝗼𝗻𝗥𝗶𝘀𝗸. Risk classification per organisation. High Risk. At Risk. Healthy. The CASE logic that defines those tiers lives in one place. Every analyst uses the same definition. Nobody builds their own version. 𝗩𝗪_𝗧𝗼𝗽𝗣𝗿𝗼𝗷𝗲𝗰𝘁𝗣𝗲𝗿𝗢𝗿𝗴. A CTE inside a View. Ranking logic hidden completely. The analyst just queries a name and gets the top project per organisation with no idea how many steps are underneath it. 𝗧𝗵𝗮𝘁 𝗹𝗮𝘀𝘁 𝗽𝗮𝘁𝘁𝗲𝗿𝗻 𝗶𝘀 𝘁𝗵𝗲 𝗺𝗼𝘀𝘁 𝗽𝗼𝘄𝗲𝗿𝗳𝘂𝗹 𝘁𝗵𝗶𝗻𝗴 𝘄𝗲 𝗰𝗼𝘃𝗲𝗿𝗲𝗱. A View can contain a CTE. All the complexity disappears. What the outside world sees is a clean name. 𝗖𝗹𝗮𝘀𝘀 𝟭𝟭 𝗱𝗼𝗻𝗲. Guided project next. Corridex Logistics. 15,000 shipment records. Real business problems. First thing we build is the foundation View the entire analysis sits on. 𝗖𝗼𝗵𝗼𝗿𝘁 𝟵 𝘀𝘁𝗮𝗿𝘁𝘀 𝗝𝘂𝗻𝗲 𝟮𝟱𝘁𝗵. 4 months. nestuge.com/mzlik606d #Datafam #SQL

most task apps made me feel behind before i even start. running 50 client accounts taught me the opposite problem: work doesn't need more speed, it needs less noise. that's the itch behind TaskFlow :) #buildInPublic