David Gadyan

David Gadyan is an Upwork EXPERT-VETTED AI Engineer and Data Scientist (AI agents, generative AI, SaaS) based in Barcelona with 8+ years of experience and 190+ delivered projects. He integrates AI into websites and products and builds everything around it: the UI/UX, the full-stack application, the database, the deployment and DevOps, the security and compliance work, and the analytics and marketing that prove the result.

Ten: AI integration for websites and products; agentic AI systems and automation; full-stack web and app development; UI/UX design and dashboards; AI ERP and SaaS development; data science, machine learning and computer vision; databases and data engineering; deployment, DevOps, AIOps and LLMOps; security and compliance (penetration testing, audits, GDPR); and AI marketing with GA4, Google Ads, SEO and GEO/AEO.

Yes, and it is the most common request. I add LLM-powered chat, search, summarisation, document extraction or recommendations to the stack you already run, whether that is Next.js, React, WordPress, Vue or a custom back end, through a FastAPI or Node service and a small, well-tested front-end component. Your existing site is not rewritten; it gains a capability.

Yes. A recent example is the S1mple.ai AI lead-scoring engine, which plugs into HubSpot, Monday and Zoho and into Postgres, MySQL, Supabase and MongoDB, with a Next.js dashboard, FastAPI back end and Kubernetes scaling. Multi-tenant access, roles, billing and reporting are part of the build, not an add-on.

Yes. I run penetration tests and security audits on web applications, APIs and AI systems (including prompt-injection and data-leakage checks on LLM features), deliver a ranked findings report with reproduction steps, and fix what I find. Data handling is designed around GDPR, with HIPAA controls where healthcare data is involved, as in the Virgil.ai health-insurance assistant.

Yes. I containerise with Docker, deploy to AWS, GCP, Vercel or Kubernetes, set up CI/CD, logging, alerting and uptime monitoring, and for AI features add LLMOps: prompt and model versioning, evaluation sets, cost and latency dashboards. You get a system that is observed, not just launched.

Yes. That includes RAG over your own documents with source citations, multi-step agents that take actions through MCP tool servers (CRM, calendar, email, databases), n8n automations and customer-facing assistants. Examples: the Virgil.ai health-insurance assistant, the Gemini + LangGraph document-review agent for a law firm, and the crypto-regulation RAG chatbot.

It depends on scope, so there is no list price. Small, well-defined pieces (a tracking audit, a security audit, a scraper, a proof-of-concept agent or an AI feature on an existing page) are typically a few days. Full products, SaaS builds and ERP modules are scoped as milestone projects. The fastest way to a real number is a short description of the problem and of the data you already have.

A proof of concept or an AI feature on an existing site is usually one to two weeks. A production system (UI, API, database, deployment, monitoring and handover) is typically four to eight weeks depending on data readiness. Audits, tracking fixes and pentests are measured in days. Data access almost always decides the timeline, not the code.

Front end: Next.js, React, React Native, TypeScript, Dash and D3. Back end: FastAPI, Flask, Node, REST and typed APIs. AI: OpenAI, Gemini and Claude APIs, LangChain, LangGraph, MCP, n8n, RAG with vector stores such as Pinecone, HuggingFace fine-tuning. Data: PostgreSQL, MySQL, MongoDB, Supabase, BigQuery, Firebase, Scrapy, Playwright. Infra: Docker, Kubernetes, AWS, GCP, Vercel, CI/CD. Marketing: GA4, GTM, Google Ads, Meta, schema.org. Languages: Python, SQL, TypeScript/JavaScript, R.

Usually one of three causes. The measurement ID on the site does not match the property you are reading. Or the event fires but was never marked as a key event in GA4, so it is stored and never counted. Or the conversion happens off your domain, a form that posts to a third party and lands the visitor on someone else's thank-you page, so no pageview of yours ever records it. All three are quick to diagnose and quick to fix. A consent banner misconfigured for Consent Mode v2 is an increasingly common fourth cause.

Yes. The audit checks whether every enquiry path produces a countable event, whether key events are marked as conversions, whether consent mode is passing data correctly, what page weight costs you in paid traffic, and what search engines and AI assistants currently say you do. Findings are ranked by payoff per hour, each with a free way to verify it yourself.

190+ projects and 50+ long-term clients, including SecondSight, Equeum, Greenwich.HR, Chilldays, KPMG, ScanLabMR, Arnotts Technology Lawyers, Happe Homes, NLOAI, linkaX and FCL Analytics, across the US, UK, EU, Australia and the Middle East. Upwork Expert-Vetted (top 1%), $500k+ earned. Reviews are in the testimonials section.

Through Upwork. Describe the problem rather than a specification: a feature that will not ship, a site that gets traffic but no enquiries, a system nobody dares to deploy. You get a straight answer on whether it is worth solving and roughly what it would take, and the whole engagement (scope, milestones and payment) runs on Upwork.

Yes. I am based in Barcelona (CET) and have worked remotely since 2018 with clients in the US, UK, Europe, Australia and the Middle East, in English, Spanish, Russian and Armenian. Clients consistently mention clear asynchronous communication and regular written updates.

Yes. You own the deliverables outright on final payment: source code, models, infrastructure configuration and documentation. NDAs are signed as a matter of course, and handover includes whatever your team needs to run, monitor and maintain the system without me.