Available for AI / LLM freelance & roles

Building production LLM systems
that take real action.

AI & Automation Engineer · LLM Agents · RAG · Tool-Calling

I design and ship agentic AI — tool-calling assistants, RAG pipelines with cited answers, and observable multi-agent workflows. Every project below is a live, clickable demo deployed on Vercel.

10+
Live AI demos shipped
2022
Building AI since
−95%
Cost cut via trace-driven tuning
Multi
Agent · RAG · automation
About

From prompt to production.

I'm Adel — an AI & Automation Engineer based in Cairo, focused on LLM systems that do more than chat: they retrieve, decide, and call tools to get work done.

I've built and deployed WhatsApp AI concierge bots, OpenAI-powered RAG pipelines, multi-agent screening systems, and n8n + GoHighLevel automation across hospitality, healthcare, e-commerce, and SaaS clients. My work pairs solid agent architecture with observability — Langfuse tracing and eval harnesses — so I can measure a pipeline and tune it: parse-failures to zero, accuracy up, cost and latency down.

I care about demos a recruiter can click. Each project ships as a real app on Vercel with a public URL.

Cairo, Egypt Arabic & English CS Graduate Freelance · 2022–Present Ex-Intern @ Maxim Investment Group
Capabilities

The AI engineering stack.

Agents & Tool-Calling

  • OpenAI function / tool calling
  • LangGraph multi-agent pipelines
  • smolagents · LlamaIndex
  • Thought → action → observation loops

RAG & Retrieval

  • Embeddings & vector search
  • pgvector / Supabase
  • Inline citations & grounding
  • Structured output parsing

Observability & Eval

  • Langfuse tracing
  • Labeled eval harnesses
  • Failure-mode analysis
  • Accuracy / latency / cost tuning

Vision & Documents

  • Vision-language models
  • Invoice / receipt / ID extraction
  • Bounding-box grounding + OCR
  • Confidence scoring

Automation

  • n8n end-to-end pipelines
  • GoHighLevel CRM workflows
  • WhatsApp Cloud API bots
  • Paymob / payment flows

Build & Deploy

  • Python · TypeScript / JS
  • Next.js · FastAPI · Gradio
  • Vercel serverless functions
  • Supabase · Redis · Celery
Live Projects

Click any demo — they're all real.

Deployed apps with public URLs. Hover a card to scroll through the full page.

LiveThastock — AI market intelligence platform

Thastock — AI Market Intelligence Platform

A full SaaS product I designed, built and shipped end to end. An AI-written brief on your watchlist at 08:30 every trading day (overnight moves, RSI, earnings, macro), an alert engine firing on price · RSI · volume · MA-cross with AND/OR compound conditions, AI research per ticker (momentum, risk score, analyst consensus, directional bias), SEC Form 4 insider filings parsed in real time, an 8,000-equity screener, portfolio optimizer and stress testing. Four live pricing tiers — free forever through $44/mo.

Next.jsOpenAIMarket data APIsSEC EDGARStripe-style tiersVercel
LiveLuca's Trattoria — AI restaurant phone host

Luca's Trattoria — AI Phone Host (orders → kitchen, tables → floor plan)

Restaurants miss up to 43% of calls in the dinner rush; 6 in 10 were orders or tables. Gia answers every call: takeout orders → POS / kitchen printer with KDS timers and a ready-text to the customer when the kitchen bumps the ticket; an 86 board so sold-out items are never offered (swaps suggested live); allergies flagged every time; upsells; totals read back and paid; reservations onto the floor plan with covers-by-the-hour and deposits for big parties. Toast / Square · OpenTable / Resy.

Vapi / RetellGPT-4o tool-callingToast · SquareOpenTable · ResyTwilio SMSFastAPI
LiveBrightside Dental — AI front desk

Brightside Dental — AI Front Desk (book · remind · refill)

1 in 5 dental appointments is a no-show ($200–375 of empty chair) and 1 in 3 new-patient calls goes unanswered. Mia books new patients with insurance verified on the call (right provider, right slot, forms texted), makes the day-before reminder calls and reschedules on the spot, offers cancellations to the waitlist within minutes, runs hygiene recall and same-day emergency triage. Week-schedule dashboard with booked-by-AI slots, confirmation ring and waitlist; Open Dental / Dentrix sync.

Vapi / RetellGPT-4o tool-callingOpen DentalInsurance eligibilityTwilio SMSFastAPI
LiveHarbor & Vine Realty — AI listing line

Harbor & Vine Realty — AI Listing Line (qualify + book showings)

The median agent answers a listing inquiry 15 hours later; 78% of buyers go with whoever answers first. This AI concierge answers in 4 s, qualifies budget · pre-approval · timeline · must-haves into a lead score, books the showing with the listing agent and texts the details, routes hot leads, nurtures the not-ready, and turns seller "what's my home worth" calls into CMA appointments. Listings board where inquiries pin onto each property; syncs to Follow Up Boss / kvCORE.

Vapi / RetellGPT-4o tool-callingFollow Up BossLead scoringTwilioFastAPI
LiveLakeside Family Law — AI visual voicemail for IP desk phones

Lakeside Family Law — AI Visual Voicemail for IP Desk Phones

80% of callers sent to voicemail hang up; the rest wait ~8 h to be heard. An AI attendant takes a structured message instead of a beep, then transcribes, summarizes, flags urgency and routes it — pushed to Cisco / Yealink / Poly desk phones through real CiscoIPPhoneMenu / CiscoIPPhoneText XML (one softkey to call back) and a web inbox with transcript, matter and assignee. Routine calls get routed, not stacked; hang-ups get an auto text-back.

IP Phone XMLWhisperGPT-4oVapi / RetellTwilioFastAPI
LiveBlue Ridge Plumbing — AI revenue recovery agent

Blue Ridge Plumbing — AI Revenue Recovery (GoHighLevel + Retell)

Plumbers miss up to 40% of calls and 85% of those callers never call back (~$13.5K/mo). This agent recovers it: missed call → text-back in 8 s, AI callback in 52 s, job booked; unsold estimates followed up on day 3, objection handled, 0% financing offered, install booked; past customers reactivated for maintenance. A plumber's dispatch desk shows trucks, a pipe-gauge recovery line, ticket-paper jobs with BOOKED stamps and the text-back log.

Retell AIGoHighLevelOutboundSMS text-backTwilioFastAPI
LiveNordFrost Appliances — AI customer-support voice agent

NordFrost Appliances — AI Customer Support Line

A human-handled support call costs ~$7.16 after a 4-minute hold. This AI support agent answers instantly and closes tier-1 calls end to end: looks up the serial and verifies warranty, troubleshoots from the product KB (error codes, resets), dispatches an authorized technician with the part pre-ordered, sells parts by phone, and detects frustration → warm-transfers to a human with the full history. Live support dashboard: containment vs 41% industry avg, $1.18/call, dispatch, parts, escalations, CSAT.

VapiGPT-4o tool-callingWarranty APIWarm transferZendesk-readyFastAPI
LiveStoreRight — AI voice receptionist for self-storage

StoreRight — AI Voice Receptionist for Self-Storage

A phone agent that answers 100% of calls in 1s, 24/7: quotes from live unit inventory, reserves the unit, books the tour and texts confirmation; takes tenant payments and enrols autopay; verifies identity and resets gate codes for locked-out tenants; auto-texts quotes to not-ready callers. Every call lands in a branded operator dashboard (leads · tenants · payments · inventory) via a Vapi tool-calling webhook. Storage facilities miss ~40% of calls (≈$58K/yr) — this misses none.

VapiGPT-4o tool-callingOpenAI TTSFastAPITwilio-readyRender
LiveMailMind — AI email triage & draft engine

MailMind — AI Email Triage & Draft Engine

Reads every incoming email, tags urgency & intent, and drafts a reply in your tone — a LangGraph pipeline (classifier → drafter → tone-critic) with human-in-the-loop approval before anything sends. Traced in Langfuse; measured 83% classification accuracy at ~$3.77 / 1,000 emails.

LangGraphGPT-4oHuman-in-the-loopLangfuseFastAPI
LiveVoxDocs — voice + RAG document assistant

VoxDocs — Talk to Your Documents (Voice + RAG)

Ask a question out loud — VoxDocs transcribes it (Whisper), retrieves the relevant passages, answers grounded only in your documents with citations (GPT-4o), and speaks it back (OpenAI TTS). An anti-hallucination guard refuses when it's not in the docs. Bring-your-own-docs; traced in Langfuse.

WhisperGPT-4oOpenAI TTSRAG + citationsLangfuseFastAPI
LiveIntakeIQ — AI voice intake & booking agent

IntakeIQ — AI Voice Intake & Booking Agent

Talk to it in your browser: a voice agent (Whisper → GPT-4o → OpenAI TTS) driven by a LangGraph state machine (greet → qualify → schedule → confirm) that qualifies the caller, captures intake, and books a real calendar invite (.ics). Every turn traced in Langfuse. Measured 100% booking completion across 10 test calls (~23s each).

WhisperGPT-4oOpenAI TTSLangGraphLangfuseFastAPI
LiveClearPath AI — agentic prior-authorization review

ClearPath AI — Agentic Prior Authorization

A LangGraph agent for healthcare prior-auth: retrieve → proposer → critic → judge, grounded in clinical criteria (RAG). Auto-resolves clear cases and escalates every denial and low-confidence case to a human, with a send-back self-correction loop. Every node traced in Langfuse. 55% auto-resolved at ~$0.005/decision.

LangGraphGPT-4oLangfuseHuman-in-the-loopRAGFastAPI
LiveTrustQuery — governed NL-to-SQL analytics agent

TrustQuery — Governed AI Data Analyst

Ask a data warehouse in plain English, get governed answers a CFO can trust. A governed semantic layer gives every metric one definition; metadata-RAG grounds the plan; the LLM emits a validated query spec (never raw SQL) that a compiler turns into SQL using only certified joins — structurally preventing fan-out inflation. Every answer is reconciled against an independent trusted-SQL report.

NL2SQLSemantic layerMetadata-RAGDuckDBEval harnessPython
LiveMulti-agent CV Screener

Multi-Agent CV Screener

A LangGraph pipeline (router → retriever → scorer) fully instrumented with Langfuse. The case study uses traces to fix real failure modes: parse-failures 100%→0%, accuracy 83%→89%, latency −37%, cost −95%.

LangGraphLangfuseOpenAIEval harnessPython
LiveRAG Document Intelligence

RAG Document Intelligence

Source-cited Q&A over contracts, reports, and PDFs: OCR-fallback ingestion, structure-preserving chunking with page metadata, hybrid retrieval (Chroma + BM25) with reranking, page-level citations, and an anti-hallucination guard that flags low confidence instead of guessing.

PythonLangChainChromaHybrid + rerankOpenAI
LiveLeadForge — Lead Enrichment & Qualification

LeadForge — Lead Enrichment & Qualification

Turns a messy Data Axle / ReferenceUSA export into enriched, evidence-backed leads: cleans the file, web-enriches each company (owner, email, LinkedIn, PE/franchise red flags with source URLs), classifies Pass / Needs Review / Reject with fit + confidence scores, and upserts into Airtable, deduped. Live upload → streaming enrichment.

PythonFastAPIOpenAI web searchAirtableData enrichment
LiveCard ID & Pricing

Card ID & Pricing

Turns a messy trading-card scan or listing title into clean canonical card data (year, brand, set, parallel, grade) via entity resolution (blocking → scoring → threshold), then prices it from sold comps. Variant-aware: the same #280 prices as Base $41 / Silver $1,775 / Red $5,200.

PythonFastAPIOpenAIEntity resolutionFuzzy matching
LiveAI Hotel Concierge

AI Hotel Concierge

An autonomous tool-calling agent that reads a guest message and takes action — answers from the knowledge base with citations, checks availability, reserves rooms, and saves contacts to a CRM, deciding which tools to call.

Next.jsOpenAI function callingAgent loopLangfuse
LiveChat with Documents

Chat With Your Documents

A real RAG app: upload PDFs → ask questions → get answers with inline page citations, grounded in your files. Per-page parsing, chunking, embedding, vector search, and token-streaming chat.

Next.jsOpenAI embeddingspgvectorSupabaseRAG
LiveVisual Document AI

Visual Document AI

Upload an invoice, receipt, or ID → structured JSON with bounding boxes drawn over every detected field, confidence scores, and a baseline (OCR+regex) vs vision-language-model comparison. Works on real photographed and handwritten docs.

Next.jsVision LLMOCR groundingStructured output
LiveNutriPlan AI

NutriPlan AI

Upload an InBody scan (read automatically with vision) or enter your stats → science-based calorie & macro targets (Mifflin-St Jeor) and a full AI-generated daily meal plan. Swap or remove any food and regenerate.

Next.jsOpenAI + VisionTargets engineVercel
LiveCV Analyzer

CV Analyzer

Upload résumés + a job profile → ranked shortlist with per-candidate fit scores, AI summaries, skill-gap analysis, and red flags — all RAG-grounded in the actual CV text.

Next.jsOpenAI + embeddingsRAGPDF parsing
LiveStocksawy

Stocksawy — EGX Intelligence

Pulls live Egyptian-market news → an LLM extracts sentiment, urgency, and per-stock BUY / SELL / WATCH signals with reasoning, plus a sector heatmap. React frontend + serverless backend.

React (Vite)OpenAIVercel serverlessGoogle News
LiveStem SaaS backend

Stem Separation SaaS

A production-style backend: upload a song → async AI stem separation job. JWT auth, quota checks, a job queue with progress, and a Pro tier — built on FastAPI + Celery + Redis with documented APIs.

FastAPICeleryRedisJWTAsync jobs
Automation & WordPress

The plumbing that runs the business.

Custom WordPress plugins and CRM automation deployed for real hospitality & healthcare clients — lead capture, AI scoring, and conversion tools.

LiveDental Implant Cost Estimator

Dental Implant Cost Estimator

A custom WordPress plugin: a multi-step calculator that gives patients a realistic treatment-cost range in under 60 seconds, then captures the lead. White-labelled and deployed across multiple dental clinics (Mill Street, Tartan, Santa, MSD, City Centre).

WordPress pluginPHPMulti-step UXLead capture
LiveDental clinic websites

Clinic & Business Websites

10+ WordPress / Elementor sites for dental clinics and businesses — conversion-focused landing pages with the estimator, GoHighLevel lead forms, and booking funnels wired in. Two live examples: City Centre Dentistry and Smiles in Waterloo.

WordPressElementorConversion UXGHL funnels

GoHighLevel Lead-Capture Plugin

  • "Contact Form + GoHighLevel" WP plugin (v2.6+)
  • Shortcode-driven, fully customisable forms
  • Auto-syncs leads to the CRM with UTM tracking
  • AI lead-scoring via ChatGPT

Clinic Review-Gating System

  • Sentiment-routes patients after a visit
  • Happy → Google Reviews; unhappy → private feedback
  • Automated SMS follow-up
  • Protects & grows public rating

n8n & Provisioning

  • 4 WP plugins that auto-provision GHL on install
  • Forms, pages, CRM fields & workflows
  • End-to-end n8n automation pipelines
  • CFT scrap-metal cost estimator plugin
Experience

Shipping AI for real clients.

2022 – Present

AI Systems Developer — Freelance

Self-employed · Egypt & global clients
  • Designed & deployed AI automation systems across hospitality, healthcare, and e-commerce.
  • Built 4 WordPress plugins that auto-provision GoHighLevel forms, pages, CRM fields, and AI workflows on install.
  • Created 20+ lead-capture forms per client with UTM tracking, linked workflows, and AI scoring via ChatGPT.
  • Built a clinic review-gating system: sentiment-routed patients to Google Reviews or private feedback with automated SMS.
Jul – Sep 2025

IT Software Engineering Intern

Maxim Investment Group — Cairo
  • Contributed to the Mega Plug EV-charging platform in Python, Java, and SQL.
  • Took part in code reviews and technical documentation.

Eva Concierge — WhatsApp AI Hotel Bot

End-to-end AI concierge for a 3-property Egyptian boutique hotel collection — booking, pricing, check-in, events, and payments in Arabic & English.

  • 6 sub-workflows behind a Master Router
  • Room availability + EGP tax/service fee logic
  • Deposits via Paymob, ID capture at check-in
  • Full audit: resolved P0 credential & error-branch bugs
n8nSupabaseOpenAIWhatsApp Cloud APIPaymob
Credentials

Certified in agentic AI.

Click a certificate to view it full-size.

DataCamp AI Engineer for Developers Associate Certificate

AI Engineer for Developers — Associate

Professional certification in building AI-powered apps with LLMs: the OpenAI API, prompt engineering, function calling, RAG, and chatbots. Credential ID AIEDA0011247893456.

DataCamp · Associate
Hugging Face AI Agents Certificate

AI Agents Course

Building autonomous LLM agents: tool use, agentic RAG, multi-step workflows, evaluation & observability across smolagents, LangGraph & LlamaIndex.

Hugging Face
DataCamp OpenAI API Certificate

Working with the OpenAI API

Chat completions, multi-turn conversations, output control (tokens/temperature), and the moderation API — hands-on in code.

DataCamp

Let's build something intelligent.

Open to AI / LLM engineering roles and freelance work. The fastest way to reach me is email.

adelsherif8@gmail.com