Full-stack & AI engineer focused on turning cutting-edge research into auditable, production-ready systems. My work is anchored in cost-aware LLM architecture, bilingual Arabic/English products, and evidence-based AI tooling.
- 🔭 I'm building ARSENAL — a master pipeline unifying 10 top-tier LLM agent systems.
- 💼 Available for freelance work → Mostaql profile. I take on RAG/SaaS builds, AI agent integrations, and Arabic-first NLP.
- 🌱 Researching Cognitive Economy — when to spend premium inference vs. cheaper tiers.
- 🗣️ Ask me about:
LLM Agents,RAG,Prompt Engineering,Next.js,FastAPI,Semitic NLP.
|
Production-ready RAG chatbot starter with Arabic/RTL, multi-document chat, and streaming responses. Next.js 16 + Vercel AI SDK + shadcn/ui. 
|
Kaggle AI Agent Security Competition — score 76.995. Multi-step attacks, WhiteRabbitNeo optimization, AgentDojo red-teaming. 
|
|
Bilingual (AR/EN) evidence-based library of 61 AI cost-reduction techniques, classified by evidence strength with MCDA scoring — no hype. 
|
Unifies 10 agent systems (APE, OPRO, ToT, LATS, Reflexion, Voyager, Meta-Prompting, Self-Refine, Prompt Report, AI Scientist v2) into one design. 
|
|
Hybrid neural/symbolic router using Scattering Degree to route Semitic words between a neural model and a symbolic morphological engine. 
|
A self-evolving intelligence platform that aggregates verbatim research from SOTA papers and synthesizes them into a cognitive mesh. 
|
|
ARSENAL-powered epistemic research kit: claim lattices, dialectic briefs, freelance scopes, and Kaggle notebook spines (L0–L6 pipeline). 
|
Diagnosing scaffolding bugs and measuring architecture value in LLM orchestration systems — empirical evaluation framework. 
|
Curated prompt/logic/flow + Mermaid graph extractions from landmark LLM papers. Active index: llm-agent-research-extractions · standalone piece: Voyager
Open to freelance gigs, collaborations on AI agent evaluation, open-source audits, and Semitic NLP. → Hire me on Mostaql · ✉️ faresrafat3@gmail.com