m
mo_ashraff2

Mohamed Ashraf

@mo_ashraff2

AI Engineer

Egipto
Árabe, Inglés
Parte de la información aparece en idioma inglés.
Sobre mí
I build AI that ships to real users — RAG systems, AI agents, and LLM chatbots that work in production, not just demos. At KBLabs I build multi-agent WhatsApp assistants for live customer conversations, grounded in real data with anti-hallucination guardrails. What I can build for you: Chat with your docs & data (RAG) AI agents & workflow automation Chatbots (web, WhatsApp, support) LLM fine-tuning & deployment Native Arabic speaker — AI that works in Arabic, not just English. Tell me what you need built and I'll tell you how I'd approach it.... Lee más

Habilidades

m
mo_ashraff2
Mohamed Ashraf
desconectado • 

Revisa mis servicios

Implementación y despliegue de IA
I will build an ai chatbot that answers from your documents using rag
DeUSD 75 / proyectoMás información
Implementación y despliegue de IA
I will build a custom ai agent that uses your tools and apis
DeUSD 75 / proyectoMás información

Porfolio

Experiencia laboral

Self_Employed

AI Engineer

Self Employed • Tiempo completo

Aug 2025 - Present • 11 mos

AI Engineer at KBLabs, building production multi-agent AI systems with Google ADK, integrated into .NET/MongoDB backends. FAHEM AI — Intelligent WhatsApp Shopping Assistant Architected a Python multi-agent orchestration system (Google ADK + LiteLLM) executing semantic product search, cart manipulation, and checkout APIs. Built 48-hour session management with active resumption and cross-session memory, persisting LLM-generated behavioral summaries to PostgreSQL. Used ADK's context compaction to dynamically summarize conversation history, keeping search accurate while cutting token costs. Containerized the webhook layer and databases with FastAPI and Docker Compose for crash-safe deployment. Estatia — Real Estate AI Concierge Built a stateless WhatsApp concierge replying to property buyers in Egyptian Arabic, English, and Franco-Arabic. Each turn runs two isolated agents: a Concierge with seven live-inventory tools, and a Profiler emitting CRM tags, summaries, and funnel stage. Answers are grounded strictly in live backend inventory rather than a stale vector store, protected by guardrail callbacks (scope enforcement, prompt-injection screening, output guard) and a deterministic layer that traces every figure back to real tool data before it reaches the customer. Includes automatic human handoff and an evaluation suite scoring the agent across 11 datasets. Educational Books RAG System Engineered a document ingestion pipeline (LangChain + Qdrant) vectorizing large textbooks into semantic chunks, enabling complex semantic querying and auto-generated student quizzes, with LangSmith for monitoring and query tracing.