AI engineering portfolio · live systems

Don’t read about my work. Run it.

Three original, interactive projects spanning data engineering, agentic RAG, model evaluation, and MLOps—each backed by a live endpoint.

Launch the labs LinkedIn ↗

PUBLIC-SAFE Original personal builds using synthetic records only. No employer code, data, or internal architecture.

portfolio://runtime

01Entity resolution APIONLINE

02Multi-agent RAGONLINE

03ML monitorONLINE

INPUT→ENDPOINT→VISIBLE OUTPUT

Interactive project lab

Three projects.
Built end to end.

Change the inputs, call the APIs, and inspect the system’s decision—not a screenshot or a GitHub link.

01

Data engineering · Knowledge graphs · Entity resolution

Atlas Resolve

A record-linkage engine that decides whether two noisy asset records describe the same real-world entity, then builds a canonical record with traceable feature contributions.

TypeScriptWeighted similarityGraph-ready outputExplainability
POST/api/resolveready
Scenario
Source A · EDR
Source B · CMDB
Decision

Run the endpoint to score these records.

Feature contribution
Canonical output
{ waiting: true }
Why it matters

Enterprise analytics fail when identity is unreliable. This demo makes the matching logic observable instead of hiding it behind a black-box join.

Production evolution

Move features to Spark, learn weights from labeled pairs, persist lineage, and emit graph edges for attack-path analytics.

Project walkthroughVideo coming soon
03:00
02

Agentic AI · RAG · Tool orchestration · LLM evaluation

Aegis RAG

A multi-agent investigation copilot that retrieves synthetic evidence, delegates analysis to specialists, cites its sources, and passes the answer through an explicit judge.

RAGMulti-agentQuery expansionLLM-as-a-Judge
POST/api/agentready

This public sandbox uses a deterministic reasoning layer over a synthetic corpus—no API keys, customer data, or hidden employer context.

Agent trace
01
PlannerWaiting for a question.
waiting
02
RetrieverSynthetic corpus ready.
waiting
03
SpecialistsIdentity and endpoint tools ready.
waiting
04
JudgeEvaluation rubric ready.
waiting
Grounded response

Run the agent to generate an evidence-backed investigation summary.

Why it matters

A useful agent must expose what it retrieved, which specialist did what, and why the final answer passed evaluation.

Production evolution

Swap in embedding retrieval and tool APIs, isolate permissions per agent, add model-based evals, and log traces for regression testing.

Project walkthroughVideo coming soon
03:00
03

Machine learning · MLOps · Drift detection · Explainability

Drift Lab

A live model-observability surface for a synthetic email-risk classifier. Stress the input distribution, tune the decision threshold, and watch model quality change.

PSI driftConfusion matrixThreshold tuningFeature impact
POST/api/monitorready
Changes sender, URL, and language feature distributions.
Higher values reduce false positives but may miss attacks.
Precision—
Recall—
F1 score—
False-positive rate—
PSI drift—WAITING
Risk-score distribution
baseline current
Confusion matrix
Pred +
Pred −
Actual +
—True positive
—False negative
Actual −
—False positive
—True negative
Global feature impact
Why it matters

A model is not done at deployment. Thresholds, drift, class imbalance, and false-positive cost decide whether it remains useful.

Production evolution

Stream prediction logs, add delayed labels, monitor slices, alert on PSI and performance, and automate champion–challenger evaluation.

Project walkthroughVideo coming soon
03:00

The engineer behind the lab

From model idea to observable system.

I’m a mechatronics engineer with an emphasis in artificial intelligence and computer vision, currently working as a Lead Cybersecurity Data Scientist.

I build across the whole path: data contracts, features, models, APIs, agent orchestration, evaluation, monitoring, and the interface where a person decides whether to trust the result.

CorePython · SQL · ML · Data engineering
Applied AIRAG · Agents · Embeddings · Evals
PlatformsAzure · Databricks · APIs · CI/CD