Blog Post
عنوان المحادثة: Zapierin 17 profiles$58High Demand...
التاريخ: 25.04.2026 - 26.04.2026
التصنيف: 💻 البرمجة وتطوير البرمجيات
إجمالي الرسائل: 3 | ياسر: 0 | M: 3
M
Hey Alex!I'm Mira — BSc AI Engineer, 5+ years in Python web scraping and browser automation. I've completed 40+ scraping projects using Playwright, Scrapy, BeautifulSoup, and advanced anti-bot techniques. My core has always been hardcore scraping — I only got into AI agents recently because that's what the market demands.---Why I'm a strong fit:- 5+ years with Playwright, Scrapy, BS4, lxml, Requests, aiohttp- Built systems scraping Amazon across 7+ country branches without API- Extracted products from AliExpress, Noon, SheIn, Trendyol- Full-stack: React + FastAPI + PostgreSQL + Docker — all production-grade- Built reusable scraping tools with GUI interfaces for long-term client usage---Screening Questions:1. Browser AI agent design?- React frontend for VIN input + results dashboard- FastAPI backend with async agent orchestration- Playwright browser layer with stealth plugins — each site gets its own profile- LLM decides which sites to hit, what selectors to use, and how to rank results- PostgreSQL for storage with full-text searchI start by mapping each target site's DOM, then build it as a plugin — making the system expandable from 5 sites to 100+.2. Challenges with eBay / CarParts?I scraped Amazon's full catalog across 7+ branches without API — handling their anti-bot, dynamic rendering, and rate limiting. eBay and CarParts are similar:- Dynamic content → Playwright handles natively- Rate limiting → rotating proxies + human-like delays- Inconsistent DOMs → site-specific selectors with LLM fallback- Bad data → multi-source cross-validation + confidence scoring3. CAPTCHA / bot detection?- Prevention first: Playwright stealth, real user-agents, human mouse movements, randomized delays- Residential proxies for IP rotation- Session management with cookies/localStorage to appear as returning user- CAPTCHA services as last resortMy Amazon system ran continuously undetected using Tor proxy rotation — I'll replicate this here.4. Part fit validation without proprietary database?- Tier 1: Extract fitment claims from the source page ("Fits 2019 Honda Accord")- Tier 2: Cross-reference VIN → vehicle specs → compare against compatibility tables on the site- Tier 3: LLM analysis of descriptions, specs, and images for fitment probability5. Similar systems I've built:- almadah.com — Live SaaS scraping 20+ e-commerce sources simultaneously, aggregating products with price comparison and unified catalog. Currently expanding to more sites with the owner. Same architecture as your project, just different industry.- Amazon Affiliate Automation — Full-stack platform scraping Amazon Bestsellers/Deals/New Releases across 7+ countries, with intelligent product scoring, auto-generated marketing designs + videos, Google Drive + Sheets integration, and a web dashboard. Available in Arabic now — happy to demo the English version on a call.- Medical Image Analysis — Trained CV models for MRI brain tumor detection. Directly applicable to your damaged part photo analysis requirement.---On OCR for part images:- Budget option: Open-source OCR (Tesseract + custom CV) — zero running costs- Accuracy option: Gemini Vision API — higher quality, ongoing API costs- I'll recommend the best path after the prototype based on your needs.---On site selection:I can scale to 100+ sites easily — but you know the auto parts landscape better. I'll build the extensible plugin architecture, you tell me which sites matter most, and I'll build the extraction profiles. This collaboration gets us to market faster.---Timeline:- Prototype: 2–4 weeks- MVP: 6–10 weeksBudget: Flexible — let's lock the feature scope, number of sites, and their structure first, then I'll give you a fair price that works for both sides.Ready to start with your paid test task. Happy to jump on a quick call to show you my previous projects live.Best,Mira
26.04.2026 00:03
M
---مرحباً،أنا ميرة — مهندسة ذكاء اصطناعي، وعندي أكتر من 5 سنوات خبرة في سكرابنج الويب وأتمتة المتصفحات باستخدام Python. نفّذت أكتر من 40 مشروع سكرابنج باستخدام Playwright وScrapy وBeautifulSoup وتقنيات متقدمة لتجاوز الحماية. شغلي الأساسي دايماً كان السكرابنج القوي — دخلت في عوامل الذكاء الاصطناعي مؤخراً لأن السوق بيطلب هيك.---ليش أنا الخيار المناسب:- أكتر من 5 سنوات خبرة بـ Playwright وScrapy وBS4 وlxml وRequests وaiohttp- بنيت أنظمة بتسحب كتالوج أمازون كامل من أكتر من 7 فروع بدون API- سحبت منتجات من AliExpress وNoon وSheIn وTrendyol- فول ستاك: React + FastAPI + PostgreSQL + Docker — كلها بروduction- بنيت أدوات سكرابنج بواجهات رسومية للعملاء للاستخدام طويل المدى---إجابات الأسئمة:1. كيف بتصمم عامل الذكاء الاصطناعي المبني على المتصفح؟- واجهة React لإدخال رقم VIN وعرض النتائج- باكند FastAPI مع تنظيم العامل بشكل async- طبقة المتصفح Playwright مع إضافات التخفي — كل موقع بملفه الخاص- الذكاء الاصطناعي بيقرر أي مواقع يفتح، أي selectors يستخدم، وكيف يرتب النتائج- PostgreSQL للتخزين مع بحث نصي كاملببدأ بدراسة DOM لكل موقع هدف، وببنيه كـ plugin — يعني النظام بيقدر يتوسع من 5 مواقع لـ 100+ بسهولة.2. شو التحديات مع eBay أو CarParts؟سحبت كتالوج أمازون كامل من أكتر من 7 فروع بدون API — بتعاملت مع مضادات البوت والمحتوى الديناميكي والتقييد. eBay وCarParts نفس التحديات:- محتوى ديناميكي → Playwright بيغطيها- تقييد الطلبات → بروكسيات دوارة + تأخير بشري- DOM متغير → selectors خاصة بكل موقع مع fallback بالذكاء الاصطناعي- بيانات سيئة → تحقق متعدد المصادر + نظام confidence scoring3. كيف بتتعامل مع CAPTCHA والكشف عن البوتات؟- الوقاية أولاً: إضافات Playwright stealth، user-agents حقيقية، حركة فأرة بشرية، تأخير عشوائي- بروكسيات سكنية لتدوير الـ IP- إدارة الجلسات مع cookies/localStorage لظهورك كمستخدم عادي- خدمات حل CAPTCHA كملجأ أخيرنظام أمازون شغّل باستمرار بدون كشف باستخدام Tor proxy rotation — بنفس الشي هون.4. كيف بتتحقق من ملاءمة القطعة بدون قاعدة بيانات خاصة؟- المستوى 1: سحب معلومات الملاءمة من صفحة المصدر ("Fits 2019 Honda Accord")- المستوى 2: مقارنة VIN → مواصفات المركبة → جداول التوافق على الموقع نفسه- المستوى 3: تحليل الذكاء الاصطناعي للوصف والمواصفات والصور لاحتمال الملاءمة5. أمثلة على أنظمة مشابهة بنيتها:- almadah.com — SaaS شغال بيسحب من أكتر من 20 مصدر تجارة إلكترونية بنفس الوقت، بيجمع منتجات مع مقارنة أسعار وكتالوج موحد. حالياً بوسّع لمواقع أكتر مع صاحب المشروع. نفس الهندسة المعمارية، بس صناعة مختلفة.- نظام Amazon Affiliate Automation — منصة فول ستاك بتسحب Bestsellers وDeals وNew Releases من أمازون عبر 7+ دول، مع نظام scoring ذكي، تصاميم تسويقية وفيديوهات تلقائية، تكامل Google Drive + Sheets، ولوحة تحكم. متاح بالعربي دلوقتي — جاهزة أعرضلك النسخة الإنجليزية بمكالمة.- تحليل صور طبية — درّبت نماذج رؤية حاسوبية للكشف عن أورام الدماغ بالـ MRI. بتنطبق مباشرة على تحليل صور القطع التالفة اللي طالبينها.---بالنسبة لـ OCR لصور القطع:- الخيار الاقتصادي: OCR مفتوح المصدر (Tesseract + CV مخصص) — صفر تكلفة تشغيل- الخيار الدقيق: Gemini Vision API — دقة أعلى، بتكلفة API مستمرة- باقترحلك الأفضل بعد البروتوتايب حسب احتياجك وميزانيتك---بالنسبة لاختيار المواقع:أقدر أوصل لـ 100+ موقع بسهولة — بس أنت بتعرف سوق قطع السيارات أكتر. أنا ببني الهندسة المعمارية القابلة للتوسع، وأنت بتقلي أي المواقع الأهم، وأنا ببني ملفات الاستخراج لكل واحد. التعاون هاد بيوصلنا للسوق بسرعة.---الجدول الزمني:- نموذج أولي: 2–4 أسابيع- MVP: 6–10 أسابيعالميزانية: مرنة — نحدد نطاق الميزات، عدد المواقع، وتعقيدها أولاً، وبعدين بعطيك سعر عادل يرضي الطرفين.جاهزة أبدأ بمهمة الاختبار المدفوعة. وجاهزة كمان لمكالمة سريعة أورجيك فيها مشاريعي السابقة مباشرة.تحياتي،ميرة
26.04.2026 00:05