R$441M
B2G TAM mapped
Sanitized healthcare market intelligence pipeline covering 64.3M units, 759 buyer CNPJs, and 26/27 states.
ACTIVE SYSTEMS:
LUMEN · AURA · VOIGHT-KAMPFF · DANEEL · KATHARSIS · CODEX
[6 of 30+]
Felipe Freitas
Engineering Meets Revenue.
Engineer by training, Revenue Architect by trade. From CAN-Bus interceptors to probabilistic S&OP engines, market data pipelines, and multi-agent LLM swarms. CAISR, LSS Black Belt, CEA ANBIMA, 5× MBA.
0
Years
0
Verticals
0
Revenue
Scroll
Recruiter brief // 60-second read
Revenue architecture with engineering teeth.
Best fit for roles that need business context, technical execution, and a calm adult in the room: Revenue Architect, RevOps/Data Lead, Solutions Architect, AI Automation Lead, AI Security/Risk Lead, or Fractional Tech Lead.
13+
years across sales, engineering, data, automation
R$300M+
revenue context handled across commercial roles
R$441M
B2G TAM mapped in sanitized market intelligence work
30+
AI/data/automation systems in the archive
CAISR
certified Artificial Intelligence Security & Risk
01
About
I've intercepted Toyota CAN-Bus frames at 3am, closed R$2M deals before lunch, and turned fragmented public data into board-level market intelligence. Same brain, different mode.
13 years across 7 verticals taught me that the best systems live where operations, data, and revenue collide. I build CAN-Bus interceptors with ESP32, design probabilistic S&OP engines for falsifiable forecasts, and orchestrate multi-agent AI swarms with equal fluency.
Current work spans confidential technology advisory, pharma market intelligence, B2G/B2B TAM engineering, AI security/risk, and local-first security experiments. The pattern is the same: replace fragile heuristics, manual rituals, and garbage CRM data with auditable systems that executives can actually trust.
Certified Artificial Intelligence Security & Risk (CAISR) by Red Team Leaders. Lean Six Sigma Black Belt with R$2.1M in documented savings. CEA ANBIMA certified. 5× MBA. My stack is Mechanical Engineering + Revenue Architecture + Data Engineering + LLM/Agentic Automation + AI Security/Risk + OpSec. If you need someone who solves problems that don't have templates yet — let's talk.
AI & Data
Infrastructure
Business
Education & Certifications
Mechanical Engineering
UFG — 2009–2016
CAISR
Red Team Leaders — Verify
LSS Black Belt
Lean Six Sigma
CEA Certified
ANBIMA
5× MBA
Big Data, CI, Finance, Econ Eng, Strategy
02
Proof Ledger
75%
Manual actions removed
Event-driven RevOps automation reduced service-order friction with n8n, CRM webhooks, Graph API, and field bot flows.
99.64%
OCR baseline preserved
Industrial OCR rollback protected a regulatory workflow while deleting bloat and restoring audit reliability.
747K
tokens curated
ECHO.2E turns a 14GB personal archive into a 747K-token long-context system that now motivates a controlled, privacy-gated contextual-attractor study.
40 / 15s
slides generated
Native COM automation renders granular executive decks without corrupting PPTX XML or wasting analyst time.
0 deps
this site ships static
No app framework, no runtime dependencies, no tracking pixels. The artifact itself is part of the proof.
CAISR
AI security credential
Verified Red Team Leaders certification covering AI security architecture, governance, risk management, and defensive practices across the AI system lifecycle. Verify credential.
03
Experience
> Technology Advisory Board Member & Fractional Tech Lead
CONFIDENTIAL TECH STARTUP (Stealth/NDA)
[Jul 2026 — Present]
Incoming advisory and fractional technical leadership under NDA. Scope covers architecture choices, C-level technical validation, logical stress-testing, shift-left technical debt prevention, LLM/data/automation governance, OpSec, hardening, and counterintelligence posture.
> Market Intelligence & Applied Data Engineering
GRUPO HALEX ISTAR FARMACÊUTICA S/A
[May 2026 — Present]
Building B2G/B2B market intelligence and data engineering systems for public and private healthcare markets. Work spans TAM modeling, competitive intelligence, public procurement extraction, ERP/public data joins, market share analysis, pricing signals, logistics geography, and GTM decision support.
> Head of RevOps & Digital Solutions Architect
Artemec Engenharia
[Jun 2025 — Apr 2026]
Led RevOps, sales intelligence, and digital solution architecture for industrial engineering services. Replaced legacy VBA/manual flows with n8n, Python, LLMs, SQL, and PowerBI systems; accelerated delivery by 30%, built zero-to-one BI operations, standardized SOPs/Scrum routines, and stabilized industrial OCR tooling for NR-13 workflows.
> Senior Master Executive — Partnerships & Large Accounts
Órigo Energia
[Nov 2024 — May 2025]
Responsible for prospecting, negotiating, and managing sales partnerships, as well as handling large-scale negotiations throughout Goiás state. Focus on BSB segments (commerce and industries) and Agribusiness sector.
> Head of Commercial & Market Intelligence
Accon Software
[Apr — Aug 2024]
Direct report to CEO. Led hiring and training of inbound, outbound, and support commercial teams. Structured PowerBI dashboards for KPI monitoring. Implemented ERP for business controls.
> Senior Commercial Executive
Getnet Adquirência (Santander)
[Nov 2022 — Nov 2023]
Managed large Middle Corporate accounts. Coordinated funnel of 16 Relationship Managers and 4 General Managers. Achieved 10% above-budget revenue and 4% growth in active client base.
> District Manager
Cappta S/A
[Mar 2021 — Oct 2022]
Led sales team of 20–25 professionals across 5 states. Achieved 300% increase in TPV and 230% growth in Net Revenue within one year. Expanded active client base by 101%.
> Senior Commercial Consultant (B2B)
Companhia Ultragaz S/A
[Jun 2013 — May 2020]
Managed key accounts in residential channel, leading the largest negotiation in the Company's Midwest region. Increased served volume by 87%. Trained 150+ sales professionals.
Experience -> systems map
Current pharma market intelligence
HELIX, Algorithmic S&OP, Leviathan, CNES Intelligence, geo-logistics.
Industrial RevOps architecture
POSV, MEMETIC, ThomasB-Vision, COM Deck Engine, Proposal Engine, Supabase/PostgreSQL operations DB.
Independent R&D / FOSS
ECHO.2E, Contextual Attractor Study, NOMAD:HANDHELD, PhonemeFree, bartleby-pad, Abandoned Uplink.
04
Portfolio
Project Explorer
Filter the work by hiring signal. Everything remains readable without JavaScript; controls only sharpen the scan.
Showing 16 projects.
01
HELIX — Governed Intelligence Platform
Governed commercial intelligence platform that turns controlled market, operational, and public-health data into auditable decision layers for strategy, territory, and execution. CLI-first Bronze / Silver / Gold pipelines, code-bound business rules, immutable release IDs, certified metrics, Gold-only Streamlit serving, and rootless deployment topology.
Case file
Problem: dashboard delusions, mutable metrics, and critical business rules hidden in presentation layers.
System: CLI-first ETL, versioned transforms, certified releases, and a fully decoupled serving layer.
Evidence: SHA-256 manifests, immutable release IDs, validation-before-publish gates, CI quality controls, and separated ETL/serving topology.
Boundary: sanitized system architecture; no company, client, licensed-source, internal-volume, or commercial-metric disclosure.
"Data without enforced lineage is just a corporate rumor."
02
ECHO.2E — Cognitive Doppelganger
Personal AI doppelganger built from a 14GB source archive and a 747K-token curated corpus spanning 2010–2026. Tri-layer cognitive architecture (CHAOS / EDITOR / SOUL), XML behavioral injection, Gemini long-context orchestration, and computational stylometry across 35+ metrics: vocabulary richness, n-gram profiles, bilingual code-switching, register contrast, and era-shifting.
Case file
Problem: blank-page friction without outsourcing voice or judgment.
System: long-context corpus, XML controls, tri-layer persona routing, stylometry metrics.
Evidence: 14GB source archive, 747K curated tokens, 35+ fidelity metrics.
Boundary: private corpus; architecture and metrics only.
"It writes like me. Sometimes better. That's the point."
03
ECHO.2e — Contextual Attractor Study
Private-by-default research instrument for measuring observable behavioral differences among three long-context LLM persona configurations under a frozen protocol. 36 SHA-locked prompts, a planned 1,080-generation confirmatory design, randomized sibling-fork collection, blinded annotation, stylometry, multilingual embeddings, and reproducibility manifests. Protocol and infrastructure are in development; no confirmatory data have been collected.
Case file
Problem: qualitative impressions cannot establish stable persona-conditioned differentiation.
System: frozen prompts and anchors, typed schemas, blinded annotation, leakage-safe classification, multilingual embeddings, Firth inference, and provenance manifests.
Evidence: 36 canonical prompts, 60 automated test cases, synthetic foundation validation, and a manual privacy publication gate.
Boundary: observable outputs only; no consciousness, genuine-personality, or internal-reasoning claims. Private corpus and raw records stay private.
"The experiment should be adversarial toward its own premise."
04
Algorithmic S&OP — Probabilistic Revenue Engine
Strictly decoupled 3-tier forecasting engine that replaces average-of-averages planning with falsifiable probability distributions. Auto-ETS / SARIMA / Prophet benchmarking, 13-month strategic forecasts, 3-month tactical signals, 80% and 95% prediction intervals, MinT reconciliation, and walk-forward validation across historical shocks.
Case file
Problem: subjective targets and moving averages hiding forecast error.
System: Auto-ETS/SARIMA/Prophet benchmarking, prediction intervals, MinT reconciliation, walk-forward validation.
Evidence: 13-month strategic horizon plus 3-month tactical layer.
Boundary: business-safe forecasting architecture; no internal demand series exposed.
"Hope is not a forecast. A falsifiable error curve is."
05
Leviathan — TAM Engineering Pipeline
Asymmetric RevOps pipeline that synthesizes market truth from unstructured public procurement, corporate registries, pricing tables, and healthcare datasets. Maps 64.3M units / R$441M B2G TAM, 759 buyer CNPJs, and 26/27 states through entity resolution, Regex/NLP normalization, DuckDB warehousing, and Power BI outputs.
Case file
Problem: granular TAM did not exist in clean commercial form.
System: procurement ingestion, entity resolution, free-text normalization, DuckDB warehouse, Power BI output.
Evidence: 64.3M units, R$441M B2G TAM, 759 buyer CNPJs, 26/27 states.
Boundary: obfuscated project name and sanitized metrics only.
"When the data doesn't exist cleanly, manufacture signal from the noise."
06
CNES Intelligence — Semantic & Geospatial Layer
Portable healthcare-market intelligence layer combining a PBIP/PBIR/TMDL semantic model, offline Parquet distribution, CNES enrichment, and QGIS territory overlays. Models 5,670 private establishments and 8,638 geocoded addresses for repeatable executive analysis, logistics, and account prioritization.
Case file
Problem: brittle one-off dashboards separated market semantics from territory and logistics context.
System: portable PBIP/PBIR/TMDL model, Parquet packs, CNES enrichment, geocoding, and QGIS overlays.
Evidence: 5,670 private establishments and 8,638 geocoded addresses modeled for repeatable analysis.
Boundary: sanitized semantic and map logic; no company data or sensitive account list exposed.
"A map is a sales strategy only when the semantic layer stops lying."
07
POSV — Event-Driven RevOps Engine
Post-sales automation engine built with n8n, CRM webhooks, Microsoft Graph, Teams, and a Telegram command center. Five asynchronous workflows reduced manual service-order actions by 75%, centralized field communication, routed high-value contracts automatically, and pushed data entry errors to zero.
Case file
Problem: post-sale operations depended on scattered manual actions and delayed visibility.
System: n8n workflows, CRM webhooks, Microsoft Graph, Teams routing, Telegram field command center.
Evidence: 75% reduction in manual service-order actions and zero data-entry error target.
Boundary: sanitized workflow topology; no customer or company payloads exposed.
"The CRM stopped being a graveyard and became a nervous system."
08
Veredas de Papel — Fixed-Income Anomaly Radar
FOSS engine for Brazilian fixed-income distress detection. BCB and IFData ingestion, critical spread and rate-jump rules, statistical divergence monitoring, typed Python discipline, and a roadmap toward Isolation Forest / DBSCAN multidimensional outlier detection.
Case file
Problem: retail investors see distressed fixed-income rates too late.
System: BCB/IFData ingestion, statistical thresholds, anomaly rules, typed Python discipline.
Evidence: public FOSS repo with roadmap for multidimensional outlier detection.
Boundary: public project; repo link available.
"When banks get desperate, the numbers scream. This listens."
09
NOMAD:HANDHELD — Offline-First Mobile Hub
FOSS Android hub for assembling a practical offline toolkit: local AI, maps, knowledge libraries, credentials, and field utilities. A schema-driven JSON catalog, on-device app detection, DataStore progress, bundled Markdown guides, and a high-contrast Field Sheet keep the experience useful without a backend. Project code is complete; release polish is under review.
Case file
Problem: capable offline tools already exist, but discovery, setup, and readiness tracking remain fragmented.
System: JSON-driven catalog, PackageManager detection, DataStore persistence, bundled onboarding, and a compact Field Sheet.
Evidence: 10 curated apps across 5 categories, with zero backend or network dependency after installation.
Boundary: independent implementation inspired by the upstream product philosophy; no fork or reused source code.
"What if the world ends and you only have a smartphone?"
10
MEMETIC v3 — Autonomous Content Swarm
n8n + LangChain multi-agent architecture for technical content operations. GOLEM, GALATEA, ARGUS, PYTHIA, HYDRA, and an error-handling layer coordinate copy, channel adaptation, visual generation, governance context, and publishing. Refactor reduced execution time by 47%, memory footprint by 60%, and context duplication by 66%.
Case file
Problem: technical content production was slow, repetitive, and compliance-sensitive.
System: n8n/LangChain swarm with specialized agents for copy, context, visuals, publishing, and error handling.
Evidence: 47% faster execution, 60% lower memory footprint, 66% less context duplication.
Boundary: sanitized agent topology; private workflows stay private.
"Brand voice is just deterministic orchestration wearing a nicer shirt."
11
KALEIDO — RevOps Intelligence Framework
Senior-role technical case built as a forensic RevOps intelligence system. Domain separation, 1,200 opportunities versus 20 audited records, R$74.8M pipeline analysis, SQL Shield for guarded Text-to-SQL, XAI churn/playbooks, and nine documented deliverables.
Case file
Problem: LLMs on dirty CRM data create confident hallucinations.
System: domain separation, SQL Shield, explainable churn risk, playbook generation, hardened data checks.
Evidence: R$74.8M pipeline analysis and nine documented deliverables in case format.
Boundary: evaluative/private repository; architecture disclosed, raw case data not disclosed.
"Dirty CRM data does not become strategy because a chatbot touched it."
12
PhonemeFree — Voice Obfuscation Engine
Local-first anti-deepfake research engine for defensive voice obfuscation. Pure C++17 DSP core with pitch shifting, formant warping, and 32-bit LFSR noise injection, packaged through Kotlin/NDK for Android and a custom ESP32-S3 USB microphone path. No cloud, no biometric leakage.
Case file
Problem: voice notes are becoming reusable biometric attack material.
System: local DSP pipeline with pitch shifting, formant warping, LFSR noise, Android and ESP32-S3 branches.
Evidence: dual-branch local-first architecture, no cloud, no biometric upload.
Boundary: defensive research; release posture controlled.
"Your voice is a credential. Stop uploading it to strangers."
13
bartleby-pad — Stealth HID MacroPad
FOSS RP2040/CircuitPython USB HID macropad for locked-down corporate environments. No wireless, no runtime config surface, and no plaintext payloads: injected strings are obfuscated with XOR + Base64 tied to the chip UID and decrypted only on that silicon.
Case file
Problem: locked-down environments turn engineers into manual typists.
System: RP2040 HID macropad with no wireless, no runtime config, and chip-bound payload obfuscation.
Evidence: public FOSS repo; payloads avoid plaintext at rest.
Boundary: public hardware/software project.
"I would prefer not to manually type boilerplate ever again."
14
ThomasB-Vision — Industrial OCR Survival
Industrial OCR rollback and lean rewrite for NR-13 compliance workflows. Preserved the 99.64% accuracy baseline, removed 433 non-essential files and 233K lines of dead code, reduced onboarding from 3 weeks to 3 days, and shipped a standalone Windows workflow for auditor-ready Excel outputs.
Case file
Problem: a bloated OCR rewrite threatened a regulatory inspection workflow.
System: rollback, lean refactor, synonym dictionary, circuit breaker, Excel auditor workflow.
Evidence: 99.64% baseline preserved, 433 files removed, 233K LOC deleted.
Boundary: sanitized industrial context; no proprietary inspection data exposed.
"Working code beats fashionable architecture when the audit is real."
15
The Abandoned Uplink — Honeypot
Psychological honeypot simulating an abandoned Windows 95 / SCADA terminal at ffreitasb.xyz. Flask backend, zero-dependency frontend, Cloudflare Worker edge classification, fake exposed assets, canary triggers, scanner tarpit behavior, and Telegram digest alerts.
Case file
Problem: automated scanners probe public infrastructure constantly.
System: fake Win95/SCADA interface, edge classification, tarpit routes, canaries, Telegram digest alerts.
Evidence: live psychological honeypot pattern and zero-dependency frontend.
Boundary: public decoy; operational internals partially withheld.
"You found a legacy machine. Actually, it found you."
16
COM Deck Engine — Executive Reporting Automation
Native Windows COM automation that turns raw territory KPIs into 40 executive PowerPoint slides in 15 seconds while preserving the original corporate design. Bypasses fragile PPTX XML manipulation, injects data through hidden alt-text placeholders, and eliminates the recurring two-hour copy/paste reporting ritual.
Case file
Problem: recurring executive reporting consumed hours of manual slide assembly.
System: native Windows COM automation, alt-text placeholders, chart/data injection, design-preserving rendering.
Evidence: 40 granular territory slides in 15 seconds.
Boundary: sanitized reporting architecture; no corporate deck assets exposed.
"The best PowerPoint work is the kind the machine does while you drink coffee."
Other systems in the archive
Want something more fun!??
The tactical TUI version lives at ffreitasb.cc
Let's
Talk.
Open for opportunities, collaborations, and conversations that push boundaries.