AnwarRezk

Software Engineer @ Silicon Mind | AI 👾

[ibo @ibothinks]

you don't need next.js + vercel + supabase anymore

- next.js → tanstack start or vinext
- vercel → workers
- supabase postgres → d1
- supabase auth → better auth on d1
- supabase storage → r2
- realtime → durable objects
- redis → kv
- background jobs → queues + cron triggers

one worker, one bill, ~$5/mo

[AnwarRezk @AnwarRezk]

@ipvpn2 Claude design with inspirations from sources like getdesign.md, https://t.co/mxablaOVRD, https://t.co/mxAOaVmqLf or any design/components library of choice

[AnwarRezk @AnwarRezk]

الskill دي من cloudflare برضه حلوة, بتخلي الـ AI coding agent يتعامل كأنه Security Auditor و يلف علي الـ codebase يطلع مشاكل الsecurity
https://t.co/4liSX3tgAn

[AnwarRezk @AnwarRezk]

من الحاجات المهمة الي لازم يتاخد ليها action ان أي حد يتابع الemails و الannouncements الي جاية من أي cloud hosting provider بتتعامل معاه بخصوص الmajor security issues الي حصلت الكام يوم الي فاتو و share الي موجود في الايميل مع اي agent يعمل scan للcodebase عشان يaddress أي مشاكل ممكن تحصل بسبب dependencies أو node version/update مثلا, و أكيد php لو بتستخدمه مفيهاش كلام

[AnwarRezk @AnwarRezk]

@hazem_samak101 أنا برضه شايف انها فعلا بتبقي علي حسب التجربة الشخصية بغض النظر عن الbenchmarks و التجارب بتاعة الarenas و كده عشان اغلبها بتبقي usecases مش هستخدمه اصلا ليها. أنا كنت جربت muse glimmer و muse spark 1.2 كانو لذاذ و سراع في حاجات زي exploration و quick bug fixes

[AnwarRezk @AnwarRezk]

يا فرج الله

[Google Antigravity @antigravity]

We’re rolling out Gemini 4 Argon in Antigravity to paid Google AI Pro subscribers today. Demand has been incredible and we've scaled capacity deliberately to get here. We’ve rolled out a separate weekly-only Argon quota for Pro on top of the existing Gemini bucket for 3.8 Flash so you have the best of Gemini for all your tasks.

[AnwarRezk @AnwarRezk]

@hazem_samak101 اه بالظبط، هستخدمه فعلا ك delegated agent يخلص التاسك و model اقوي زي astra او opus 5.5 يراجع عليه. مستغرب ال benchmarks بتاعته كنت متوقعه هيأدي 😅

[Relomote @Relomote]

450,000+ remote and hybrid jobs in one place. 100% free, updated every hour.

[lauren @poteto]

this has become one of my most used prompts recently:

> restate in your own words what you think my goals are and what the problem i'm trying to solve is

[Rohit Ghumare @ghumare64]

As an AI Engineer. Please learn:

> Harness engineering, not just prompt engineering
> Context engineering, not just long prompts
> Prompt caching vs. semantic caching tradeoffs
> KV cache management, eviction, reuse, and memory pressure at scale
> Prefill vs. decode latency and why they optimize differently
> Continuous batching, paged attention, and throughput optimization
> Speculative decoding vs. quantization vs. distillation tradeoffs
> INT8, INT4, FP8, AWQ, GPTQ, and when quantization hurts quality
> Structured output failures, schema validation, repair loops, and fallback chains
> Function calling reliability, tool contracts, argument validation, and idempotency
> Agent guardrails, loop budgets, tool budgets, and termination conditions
> Model routing, graceful fallback logic, and degraded-mode UX
> RAG architecture: chunking, embeddings, hybrid search, reranking, and freshness
> Retrieval evals: recall, precision, grounding, attribution, and citation quality
> Evals: golden sets, regression tests, adversarial tests, LLM-as-judge, and human evals
> LLM observability as a first-class discipline: traces, spans, tokens, latency, errors, and drift
> Cost attribution per feature, workflow, tenant, and user journey, not just per model
> Safety engineering: prompt injection defense, data leakage prevention, and permission boundaries
> Multi-tenant isolation, cache safety, and cross-user context contamination prevention
> Fine-tuning vs. in-context learning vs. RAG vs. distillation, and when each is the wrong tool
> Latency, quality, cost, and reliability tradeoffs across the full inference stack
> Production failure modes: hallucinated tool calls, malformed JSON, stale retrieval, runaway agents, and silent eval regressions
> Shipping LLM systems as reliable infrastructure, not demos wrapped around prompts

https://t.co/Q2FHIjahSx

[Mohamed Mostafa @momostafas]

للناس اللي بيدوروا على شغل زى حالاتى , ال repo دى جامدة جدا جدا بت login فيها ب claude وبيعمل job search , mass applications على حسب ال cv بتاعك وبيعملك كل ال documents اللى بتقدم على اي شغلانة ويقدملك بيها ف بصراحه انصح بيه جدا ل mass apply

https://t.co/E1zQg6QXUk

[hadzz @uns33nclu3]

وتدفعوا ليه؟ خدوا access لكل ال public library بتاع O’Reilly ببلاش:

https://t.co/xlc3qu6yk9

[Ahmed @ahmd3ssam]

مسابقة للناس المهتمين بالـAI ، بجد خش و حاول، مش هتخسر حاجة تدريب ببلاش و لو نجحت (الجائزة الآوي ٣٧ الف دولار، التاني ١٨ الف دولار، الثالث ١٠ الاف دولار )
Google - The Gemma 4 Developer Agent Competition
https://t.co/Ncjcj72fN9

[AnwarRezk @AnwarRezk]

@itamarbengvir

[Matt Pocock @mattpocockuk]

Stop caring so much about model releases.

Focus on the harness, and improving the environment your agent operates in.

You'll find yourself far ahead of the curve.

[ibo @ibothinks]

everyone posting the old oracle free vps but

4 ocpu / 24gb got cut to this in june:

- 2 ocpu (arm, ampere)
- 12 gb ram
- 200 gb disk
- always free, no card charge

still more than a $20/month hetzner box

[AnwarRezk @AnwarRezk]

@thsottiaux Add dictation here

[AnwarRezk @AnwarRezk]

Worth considering, no wonder why chatgpt sites infra is backed by cloudflare

[ibo @ibothinks]

my 5 production apps cost me $6.00 last month

- workers (compute)
- d1 (sql db)
- r2 (storage, zero egress)
- kv (cache)
- queues (jobs)
- cron triggers (schedules)

same setup on vercel + supabase + upstash is ~$200

stop overengineering your infra

[Suraj Sharma @suraj_sharma14]

As an AI Engineer, you must build these projects.

Systems that prove you can retrieve, reason & ship.

1.) Production RAG Pipeline
Build: Ingestion → chunking → hybrid search → reranking → cited answers over 1000+ docs.
Why: Retrieval quality decides if your LLM product lives or dies.

2.) Structured Output Engine
Build: Pydantic schema enforcement with retries, validation, graceful fallbacks on malformed JSON.
Why: Apps break on free text. Schemas make LLM output usable.

3.) Context Assembly Service
Build: Dynamic context builder that budgets tokens across memory, docs & tools per request.
Why: Context engineering beats prompt tweaking in production.

4.) LLM Evaluation Harness
Build: Golden dataset + LLM-as-a-judge + CI gate that blocks merges on quality regression.
Why: You cannot ship what you cannot measure.

5.) Semantic Cache Layer
Build: Embedding-based cache returning stored answers for similar queries, with hit-rate tracking.
Why: The cheapest token is the one you never generate.

6.) Model Routing Gateway
Build: Complexity-based router across 3 model tiers with fallbacks and per-request cost tracking.
Why: Unit economics decide if your AI product survives.

7.) Multi-Tenant LLM API
Build: Per-tenant keys, rate limits, token budgets, isolated data access.
Why: Enterprise customers demand isolation and spend control.

8.) Fine-Tuning Pipeline
Build: LoRA/QLoRA fine-tune of a small model on synthetic data, benchmarked vs prompt-only baseline.
Why: Know when adaptation beats prompting, and when it doesn't.

9.) Agent Memory System
Build: Working + episodic + semantic memory with compression and eviction policies.
Why: Agents without memory fail by turn five.

10.) Guardrails Middleware
Build: Input/output filtering, PII redaction, injection detection as a reusable service layer.
Why: One leak destroys trust. Safety must be infrastructure.

11.) Streaming Response Infrastructure
Build: SSE streaming with backpressure, reconnects, time-to-first-token tracking.
Why: Perceived latency is the UX of AI products.

12.) Prompt Versioning and A/B System
Build: Registry for prompts and configs with rollback, traffic splitting, outcome tracking.
Why: Prompts are production code. Treat them that way.

13.) LLM Observability Stack
Build: Tracing for prompts, tokens, latency, cost per request with alerting on anomalies.
Why: You cannot debug what you cannot see, especially non-determinism.

14.) Tool-Calling Framework
Build: Typed function schemas with discovery, retries, sandboxed execution.
Why: Tools turn chatbots into systems that actually do work.

15.) Self-Correcting RAG Agent
Build: Query rewriting, retrieval critique, confidence-based fallback to web search.
Why: Production retrieval must repair itself, not hallucinate.

Most people watch tutorials. Builders ship systems.

Bookmark & Repost.