Miguel Sánchez Durán · full-stack AI engineer · Dubai

I build AI you can actually audit.

buildingsecure ai systems, rag + evidence, ai eval harnesses, full-stack platforms, agentic workflows

AI you can put in front of real decisions: it reasons over retrieved sources, shows its working, and knows when to decline.

Get in touchLinkedIn
  • 15+ years engineering
  • Angular / Node / Python
  • RAG + evals
  • local-first AI
  • Dubai
About

How I work.

Fifteen years an engineer; the last few years spent moving deep into production AI. I use LLMs as core components, not just coding assistants. I own the architecture and the integration calls, and let AI accelerate the build. I'd rather ship something careful than something flashy.

Angular · TypeScript · Node · GraphQL · MongoDB
RAG · ONNX classify · ReAct · cite-or-refuse
local models · immunity gate · scorecards · rollback
MCP · hooks · cross-repo · automation
Signature systems

Three systems, in the field.

Three production systems I architected and shipped, built to hold up under real constraints. Described at an architectural level on purpose: the underlying work sits in sensitive, private-domain environments, so there are no client names, datasets, or internal specifics here.

Prefer to see the behaviour? Walk through a synthetic decision
in production

The Careful Machine

An AI research system for sensitive, evidence-heavy work: it routes each question through retrieval and reasoning, assembles findings from many sources, and checks every claim against its record before anything surfaces. When support is thin, it declines.

  • retrieval + reasoning
  • evidence assembly
  • cite-or-refuse
  • audit + replay
  1. Constraintsensitive, high-stakes evidence; answers must be defensible
  2. ArchitectureRAG, ONNX classification, ReAct reasoning, NER, citation verification
  3. Validationfaithfulness checks against cited sources, refusal under weak retrieval, human sign-off
  4. Resultevidence-backed reporting across many data sources
Deep dive
internal tool

The Training Forge

A gated, human-audited loop for improving local models: corrections enter as candidate lessons, candidates are screened and scored, and a new version ships only if it beats the one it replaces. Rollback is always one step away.

  • candidate lessons
  • immunity gate
  • scorecards
  • rollback
  1. Constraintimprove local models without letting bad lessons contaminate the system
  2. Architectureimmunity gate, faithfulness and competence scorecards, promotion checks
  3. Validationpromote only if the candidate beats the current model; rollback available
  4. Resulta safer local improvement loop behind the wall
Deep dive
daily driver

The Build Machine

The toolchain behind the work: hooks, agentic commands, and quality gates wrapped around the model, so one engineer can explore, change, and verify large codebases at speed, under review.

  • MCP / hooks
  • agentic commands
  • quality gates
  • automation
  1. Constraintship faster without losing architectural control
  2. ArchitectureClaude tooling, hooks, MCP servers, command layers, cross-repo mapping
  3. Validationworkflow guardrails, review passes, test and verification loops
  4. Resultfaster delivery with more repeatable engineering workflows
Deep dive
Stack

Tools I reach for.

A focused toolkit, grouped by where it sits in the system.

Frontend

  • Angular
  • TypeScript
  • React
  • RxJS
  • NgRx
  • D3 / Leaflet

Backend

  • Node.js
  • Python
  • Java
  • GraphQL
  • REST
  • SQL / MongoDB

AI systems

  • LLM orchestration
  • RAG · ReAct
  • Vector search
  • ONNX / NER
  • MCP / agents

Delivery

  • CI/CD
  • Docker
  • Quality hooks
  • Secure / air-gapped
Experience

Fifteen years. Recent work first.

Weighted to current AI and full-stack work; earlier roles in brief.

  1. Senior Software Engineer, AI / full-stack

    Datafusion Systems·Since 2024 · Dubai, UAE

    • Built sensitive, evidence-heavy AI workflows end to end: retrieval and entity extraction that turn messy multi-source data into reviewable, cited outputs.
    • Designed the orchestration and promotion gates that keep a person in control, and the AI-augmented workflow that speeds delivery without loosening review.

  2. Senior JavaScript Engineer

    Independent / consulting·2017–2024 · UK · USA · EMEA

    • Full-stack web platforms for government, healthtech, and e-commerce clients, built for regulated, sensitive-data environments.
    • Started bringing AI into client work: NLP and ML for search relevance, entity extraction, and document processing.
    • Built early LLM-assisted features and retrieval pipelines as the tooling matured.

  3. Earlier engineering roles

    Formedix · Ideagen · SlashMobility·2015–2017 · UK / Spain

    • Team-lead and engineering roles across clinical-metadata, GRC compliance, and cross-platform mobile.
Education

Where it started.

Universidad de Sevilla

Seville, Spain

  • MSc, Web & Mobile Application Development2011–2012
  • BSc, Computer Software Engineering2007–2011
Contact

Let's build something real.

Always happy to talk AI engineering, compare notes, or dig into a hard problem. Email is the fastest way to reach me, and I read everything.

contact@miguelsanchez.co.ukDownload CVLinkedIn