Analytics Engineer · Data & AI Systems · BR → CA

Engineering the pipelines, agents, and dashboards behind executive decisions.

Five years inside the engineering and technology division of a major telecom operator, building the data systems that managers, directors, and C-level leaders bet capital and headcount on. Now extending that practice into AI agents — the next interface for analytics.

5yrs
Senior data work
1.5M+
Customers served
6
Active certifications
2027
NBCC, New Brunswick
§01 · Brief

Senior data work,
built to be used.

I work as a Senior Data Analyst / Analytics Engineer inside the engineering and technology division of a major telecommunications operator. My remit covers the full data lifecycle — extraction, lakehouse, warehouse, gold semantic layer, and the Power BI surfaces leadership actually opens on a Monday morning.

The systems I build are not exploratory notebooks. They are decision infrastructure: the source of truth that managers, directors, and executives use to allocate capital, prioritize work, and steer the network.

More recently I've moved into AI agents — designing the retrieval, planning, and tool-use patterns that put the same data behind a conversational interface. I am relocating to Canada in 2027 to study at NBCC and build my career there.

§02 · Practice

Two disciplines,
one substrate.

The data platform is the same. What changes is the interface — a dashboard for a director, an agent for a field technician.

P.01 — Analytics Engineering

The pipeline from raw signal to executive dashboard.

End-to-end medallion architectures in the Microsoft Fabric and Databricks stack: source extraction, bronze landing, silver conformance, gold semantic modeling, and the Power BI report layer the business actually consumes.

  • Microsoft Fabric · Lakehouse, Warehouse, Eventhouse
  • Databricks · PySpark transformations, Delta Lake
  • Power BI · DAX, semantic modeling, governed report layer
  • Power Platform · Automate, Apps, Pages
P.02 — AI Agents

The same data, behind a conversational interface.

Building agents that decompose business questions, retrieve from governed semantic layers, call tools against operational systems, and return executive-ready answers. Grounded, auditable, and bound to the same data contracts as the dashboards.

  • Azure AI Foundry · Copilot Studio · Semantic Kernel
  • Multi-agent orchestration · planner / retriever / narrator patterns
  • RAG over Fabric OneLake and governed Power BI semantic models
  • Tool use against operational APIs and warehouses
§03 · Selected Work · 2022 – 2026

Six projects.
Six decisions they unlocked.

On anonymization Projects in this catalog are anonymized under NDA. Architectures, technologies, and patterns are accurate to the work delivered; client identifiers and exact figures are abstracted to illustrative ranges.
P-01 / 2023

Network Performance Intelligence Platform

Telecom · Network Operations

Fifty-plus network KPIs scattered across a dozen operational systems. Executive leadership lacked a unified near-real-time view of network health, and weekly steering meetings opened with thirty minutes of reconciling conflicting numbers.

Microsoft Fabric Databricks PySpark DAX Power BI Premium
SOURCE SYSTEMS OSS · network BSS · billing Field tickets SNMP telemetry CRM cases + 7 more systems BRONZE Lakehouse raw ingestion no transformation parquet · delta ~100M rows / day retention: 90d SILVER Warehouse conformed deduplicated SCD type 2 PySpark + SQL quality rules: 240+ tested · documented GOLD Semantic layer business-aligned KPI definitions role-based access star schema · DAX 52 KPIs single source of truth CONSUMPTION Executive cockpit Power BI · daily Ops dashboards Power BI · hourly Alerting Power Automate Analyst sandbox Fabric notebooks END-TO-END LATENCY · raw → gold ≈ 18 min REFRESH CADENCE Daily · incremental ADOPTION Network Operations leadership DATA CONTRACT Versioned · governed
FIG · P-01 Medallion architecture · 12 source systems consolidated into a single governed semantic layer.
12 → 1
Source systems
consolidated
−85%
Time to insight
vs. prior process
52
KPIs unified
under one contract
100M+
Rows refreshed
daily, incremental
P-02 / 2025

Field Operations Conversational Agent

Telecom · Field Service Operations

Field technicians lost minutes per incident consulting fragmented procedure manuals, network topology references, and ticket histories. Junior technicians over-relied on a small group of senior staff, creating a knowledge bottleneck.

Copilot Studio Azure AI Foundry Fabric OneLake RAG Power Platform
Technician field · mobile AGENT CORE · Copilot Studio Planner decomposes intent · routes Retriever vector search · grounded Tool layer topology API · ticket API grounded · auditable · cited KNOWLEDGE Procedure docs 3,200+ articles · versioned Network topology live · OneLake mirror Ticket history 5y · resolution patterns Asset inventory CMDB · governed EMBEDDINGS · refreshed nightly Azure AI Search · vector index RESPONSE LATENCY ≈ 4s P50 GROUNDING RATE 96% · cited DEPLOYMENT Teams · mobile · field tablets
FIG · P-02 Retrieval-grounded agent · planner / retriever / tool layer · governed knowledge backbone.
−40%
Resolution time
tier-1 incidents
−60%
Senior tech
consultations
96%
Answers grounded
and cited
800+
Field technicians
onboarded
P-03 / 2024

Churn Probability & Customer Lifetime Value Pipeline

Telecom · B2C Retention

Retention campaigns were broad and expensive — discount offers sent to the entire at-risk base without differentiation. The business needed precision: who is actually likely to leave, who is worth keeping, and what offer fits.

Databricks MLflow XGBoost Feature Store Fabric Pipelines
FEATURE STORE · governed Customer features 240+ features — usage patterns — billing signals — support history — network experience — tenure curves point-in-time correct no target leakage TRAINING · Databricks Churn model XGBoost · AUC 0.86 CLV model gradient boosted MLflow registry versioned · governed SCORING & SEGMENTATION 2×2 risk × value grid low risk high value nurture high risk high value intervene low risk low value monitor high risk low value deprioritize refreshed weekly ACTIVATION CRM cohorts offer-matched Campaign tool via Fabric pipelines A/B framework causal lift tracked Power BI · outcomes CAMPAIGN EFFICIENCY (TARGETED) ~3× lift FEATURE COUNT 240+ · versioned REFRESH Weekly · auditable
FIG · P-03 Feature store → MLflow registry → 2×2 segmentation grid → campaign activation.
~3×
Retention campaign
efficiency lift
0.86
Churn model
AUC, holdout
240+
Engineered features
governed
weekly
Scoring cadence
automated
P-04 / 2024

Executive Decision Cockpit · Infrastructure Capital

Telecom · Strategic Planning · C-Level

Annual infrastructure capital decisions — running into nine figures — were informed by static monthly reports that lagged operational reality by weeks. The steering committee needed a live decision surface with scenario simulation.

Power BI Premium Fabric Warehouse DAX (advanced) Power Apps Power Automate
LIVE INPUTS Demand forecasts region · 18-month horizon Asset utilization per cell · per route Cost actuals capex + opex Revenue mix by segment · by region SIMULATION ENGINE DAX · what-if parameters Base case committed plan Accelerated rollout +15% capex · earlier revenue Deferred −20% capex · risk modeled Region-rebalanced capital shift by demand DECISION SURFACE NPV by scenario 5-year cash flow live ranking Risk-adjusted return stochastic band Sensitivity grid tornado · top 10 drivers Steering committee view Power BI · Teams-embedded Approval flow Power Automate — scenario lock — sign-off route — audit trail — decision archive SOX-compatible DECISION CYCLE monthly → weekly SCENARIOS / SESSION 4 standardized · n custom AUDIENCE Executive Steering Committee
FIG · P-04 Live inputs → scenario engine → decision surface, with auditable approval flow for executive sign-off.
monthly → weekly
Capital decision
cycle compressed
4 + n
Standard + ad-hoc
scenarios per session
C-level
Adopted by
Steering Committee
SOX
Audit-trail
compatible
P-05 / 2023

Regulatory Reporting Automation

Telecom · Compliance & Regulatory Affairs

Quarterly regulatory submissions consumed eighty-plus analyst-hours of manual assembly: pulling from operational systems, applying validation rules, formatting to a strict regulator schema, and routing for sign-off. Any error meant rework under deadline pressure.

Microsoft Fabric Power Automate Paginated Reports Python SharePoint
TRIGGER Quarterly scheduled EXTRACTION · Fabric pipelines OSS extract BSS extract Customer registry Service catalog VALIDATION ENGINE Rule set 187 rules — completeness — referential — range checks — schema conformity Discrepancy log — auto-routed — assignee + SLA zero-tolerance gate FORMAT & ASSEMBLE Regulator schema strict XML / XBRL Python templating Paginated PDF cover · annex signed package REVIEW Compliance sign-off SUBMIT portal API Archive 7y retention EFFORT · per submission 80h → ~4h review VALIDATION RULES 187 · automated INTEGRITY INCIDENTS 0 · last 6 quarters SLA · WORST QUARTER Met with 9-day buffer
FIG · P-05 Scheduled trigger → multi-source extract → 187-rule validation → regulator-format submission with audit trail.
80h → 4h
Analyst effort
per quarter
187
Validation rules
automated
0
Integrity incidents
over 6 quarters
7y
Auditable
retention archive
P-06 / 2026

Multi-Agent Workflow for Analytical Self-Service

Internal · Analytics & Decision Support

Business users requested ad-hoc analyses faster than the analytics team could deliver, while the team itself was pulled away from strategic work to triage the queue. The need: a self-service layer that could answer routine business questions against the governed semantic model, with executive-grade narrative.

Semantic Kernel Azure OpenAI LangGraph Power BI semantic OneLake APIs
QUESTION Business user natural language 01 · PLANNER Decompose intent — what is asked? — which domain? — what time scope? — what grain? Route plan — retrieval steps — tool calls — validation checks Guardrails — scope · policy — PII redaction — refuse-out-of-domain → Retriever 02 · RETRIEVER Semantic query — DAX against gold — governed measures — RLS honored Document retrieval — policy & methodology — prior reports Tool calls — OneLake REST — Power BI XMLA — operational APIs Evidence ledger — numbers + sources — citations attached → Narrator 03 · NARRATOR Compose answer — executive register — quantified claims — each cited Self-critique — numerical check — claim ↔ evidence — hedge if uncertain Format — summary first — table on request — Power BI link Handoff — escalate if ambiguous → Analyst review OUT Answer + citations + Power BI + confidence feedback · replan on low confidence
FIG · P-06 Planner → Retriever → Narrator. Each agent has explicit role, guardrails, and a handoff contract; low-confidence answers replan.
~70%
Routine ad-hoc
now self-serve
100%
Answers carry
citations + Power BI link
RLS
Row-level security
honored end-to-end
analyst
Team refocused
on strategic work
§04 · Stack

An index
of the tools.

01 Data Platform
  • Microsoft Fabric daily
  • OneLake · Lakehouse · Warehouse daily
  • Databricks · Delta Lake weekly
  • Eventhouse · KQL monthly
02 Languages
  • SQL · T-SQL · KQL daily
  • Python · PySpark daily
  • DAX daily
  • M (Power Query) weekly
03 Analytics & ML
  • Power BI Premium · semantic models daily
  • MLflow · model registry weekly
  • scikit-learn · XGBoost weekly
  • Power BI Paginated Reports monthly
04 AI Agents
  • Azure AI Foundry active
  • Copilot Studio active
  • Semantic Kernel · LangGraph active
  • Azure OpenAI · embeddings active
05 Automation
  • Power Automate daily
  • Power Apps weekly
  • Azure DevOps · Git daily
  • Fabric Data Pipelines daily
06 Method
  • Medallion architecture default
  • Data contracts · governed semantic layer default
  • Lean Six Sigma in training
  • Stakeholder & executive comms continuous
§05 · Trajectory

From the operator,
to New Brunswick.

A deliberate path. Senior data work, then certifications that match the next decade of the field, then formal study in Canada.

2022 — Present

Senior Data Analyst / Analytics Engineer

Engineering & Technology division of a major telecommunications operator. End-to-end ownership from raw data to executive consumption.

2024 — Present

Specialization · AI Agents

Hands-on projects building grounded, multi-agent systems over the same governed data layer. Production deployments in field operations and internal analytics.

2027 →

NBCC · Information Technology: Business & Advanced Analytics

New Brunswick Community College, Canada. Formal pathway into the Canadian market, leading toward permanent residency.

Certifications
Certified · verifiable
AZ-900
Microsoft Certified: Azure Fundamentals Verify credential
Certified
Roadmap · 2026 – 2027
DP-600
Fabric Analytics Engineer Associate
In progress
DP-700
Fabric Data Engineer Associate
In progress
AB-620
AI Agent Builder Associate
In progress
LSS
Lean Six Sigma · Green Belt
In progress
AB-730
AI Business Professional
Planned
AB-731
AI Transformation Leader
Planned
AI-300
Designing & Implementing AI Solutions
Planned
§06 · Contact

If the brief sounds right, let's talk.

Email
gustavo.resendeh@gmail.com
LinkedIn
linkedin.com/in/gustavo-resende
Based
Brazil · relocating to Canada, 2027