Francisco MinguezData & AI

Capability Atlas

What technical depth exists underneath these engagements?

What you are looking at

A Capability Atlas for three primary domains: RAG & Retrieval, Applied ML, and Data & Pipelines, with labelled dependencies. It is not a skills grid.

What you can do

Read each domain at equal visual weight, open one relational exploration to see labelled dependencies, then follow a decide path or published evidence.

What it means

Engagements sit on technical depth. RAG is flagship by available depth and evidence, not by eating the visual hierarchy. Relationships stay explicit and labelled.

Primary domains

Three domains, balanced hierarchy. RAG carries more published depth and evidence availability; Applied ML and Data & Pipelines remain first-class peers.

RAG & Retrieval

Flagship by depth / evidence availability

Retrieval-backed systems: corpus, ranking, grounding, evaluation, and production-trust investigation. Chat demos are not proof.

  • Reliability investigation across query, retrieval, context, generation and end-to-end
  • Evaluation and failure taxonomy before shipping narratives
  • Published study methodology as classified evidence (not live client proof)

Depth & evidence availability · Highest published depth on this site today (study + synthetic method demos). Flagship status is evidentiary, not visual monopoly.

Applied ML

Peer domain: judgment over model theatre

Decision systems that earn a model only when they beat a simpler baseline or rule, with validation and failure cost named first.

  • Baseline-before-model judgment
  • Workload-bound model/configuration choice under explicit constraints
  • Release and regression thinking when behaviour changes

Depth & evidence availability · Synthetic method demos for choice and protection; academic ML background as classified study material, not client ROI.

Data & Pipelines

Peer domain: foundations for downstream AI

Owned data movement, quality controls, and operable pipelines that make downstream ML and RAG diagnosable.

  • Pipeline reliability and replay thinking
  • Quality and ownership as production constraints
  • Professional delivery evidence kept classified and anonymised where published

Depth & evidence availability · Professional pipeline delivery and synthetic data-quality lab material, classified; not a guarantee of your estate.

Labelled relationships

Dependencies are written as records, not a graph you must interpret. Mobile keeps the same idea as explicit domain and dependency records.

  • RAG → Data & Pipelines

    RAG reliability depends on evaluation design and data quality, not retrieval polish alone.

  • Applied ML → Data & Pipelines

    Applied ML depends on data foundations and validation before a model earns its place.

  • Data & Pipelines → RAG

    Pipeline reliability supports downstream RAG and ML by making inputs replayable and owned.

  • Data & Pipelines → Applied ML

    Stable pipelines and features keep Applied ML diagnosable when production behaviour shifts.

  • Applied ML → RAG

    Applied ML judgment (baselines, thresholds, failure cost) strengthens RAG evaluation and release decisions.

Relational exploration

One exploration interaction: focus a domain to see what it depends on and what it supports. Core meaning remains readable without JavaScript.

Focused domain

Focused domain

RAG & Retrieval

Depth & evidence availability: Highest published depth on this site today (study + synthetic method demos). Flagship status is evidentiary, not visual monopoly.

Depends on

  • Evaluation design and failure taxonomy
  • Corpus freshness, authority, and data quality
  • Clear release / trust criteria, not demo fluency

Supports

  • Production-trust decisions for retrieval-backed systems
  • Grounding and abstain behaviour under risk

Open Inspect / Trust

Focused domain

Applied ML

Depth & evidence availability: Synthetic method demos for choice and protection; academic ML background as classified study material, not client ROI.

Depends on

  • Data foundations and validation harnesses
  • A simpler baseline or rule that the model must beat
  • Named decision logic and failure cost

Supports

  • Workload-bound model or configuration choice
  • Regression-aware change judgment

Open Compare / Choose

Focused domain

Data & Pipelines

Depth & evidence availability: Professional pipeline delivery and synthetic data-quality lab material, classified; not a guarantee of your estate.

Depends on

  • Source ownership and acceptance criteria
  • Replayable transforms and operable failure handling

Supports

  • Downstream RAG reliability investigation
  • Downstream Applied ML that can be diagnosed rather than guessed

Open Build / Improve

Natural next step

Enter the decide path that matches your uncertainty, inspect published evidence with its classification, or discuss a concrete constraint.