$ whoami
Marcus N. Gabrich
Systems Architect · Tech Lead
I architect revenue-grade microservices and local-first AI infrastructure — and run both in production.
● Open to consulting & advisory work BSc Computer Science — PUC Minas SSCAD 2024 author 3rd place — Ambev national hackathon 🇧🇷 Brazil · remote
latest correction The filter learned the summariser
— published next to the result it replaced.
about
Systems Architect, Tech Lead, and Computer Scientist specializing in scalable cloud
microservices, high-concurrency revenue attribution engines, and local-first distributed
AI systems.
At RoiGenius, I operate across dual technical and executive roles: leading core
engineering while directing product roadmap, corporate cash flow, and cloud budgeting
with General Manager scope. We built a resilient 8-microservice platform integrating 89
ad and checkout platforms with zero-loss string-based currency precision (BRL).
Independently, I advise Assiny (checkout & payments) as a technology consultant.
Beyond SaaS, I architect high-performance computing on self-hosted infrastructure:
a local-first multi-agent AI platform with heterogeneous distributed compute,
bi-temporal semantic memory at scale, and a deterministic governance layer of
modular skills and security hooks. My HPC research includes a published paper
on hybrid quantum-classical neural networks (SSCAD 2024).
current research
- Calibrated decision layers — Models that answer bounded questions with probabilities instead of generating prose — and how to verify that their confidence means something.
- Benchmark methodology — Pre-registered bars, anti-contamination, honest repetition and measurements that block a run instead of describing it.
selected work
Research · S.A.C.I. · S.A.C.O.
Sep 2026 — Present
Sistema Automático de Caminhos Inteligentes — a decision layer that turns program state into routed, calibrated choices, and abstains when it should.
- Typed questions with bounded answer spaces; every answer carries a calibrated probability and a confidence band (auto / human / defer)
- Best-calibrated question at ECE 0.019; the configuration that missed the pre-registered bar is published as rejected
- Fair comparisons enforced structurally: same items per question, reference numbers never mixed with measured ones
CalibrationConformal predictionBenchmarkingPython
Quantitative Systems
Dec 2025 — Present
Proprietary execution and market-monitoring algorithms on MetaTrader 5.
- Automated execution algorithms with programmed risk-containment locks
- Ran on a demo account until a full reset in October 2026; a challenger was judged only after 10 days and 40 trades beside the current best configuration
PythonMQL5MetaTrader 5Linux
SaaS · Revenue Attribution
Aug 2024 — Present
Multi-tenant revenue attribution and marketing analytics platform running in production.
- 8 NestJS microservices on Google Cloud Run + Cloud SQL PostgreSQL 15
- Webhooks from 89 advertising platforms and payment gateways
- String-based BRL arithmetic eliminates rounding drift; DLQ (1h→16h) guarantees zero event loss
TypeScriptNestJS 11Cloud RunPostgreSQL 15RedisReact 18Cloudflare Workers
AI Infrastructure · Zero-cloud
Aug 2024 — Present
Self-hosted, zero-cloud-dependency platform for running LLM agents — private infrastructure by design.
- Heterogeneous distributed compute with GPU-accelerated workers for reranking and local inference — fully offline
- Bi-temporal semantic memory at scale (pgvector, valid_from/valid_to)
- Governance layer: model-agnostic skill modules behind deterministic security hooks
Distributed ComputeGPULocal InferencepgvectorBash
Research · SSCAD 2024
2023 — published 23 Oct 2024
HPC paper at SSCAD 2024, the Brazilian symposium on high-performance computing systems.
- A classical and two hybrid quantum-classical networks compared on the same measures (TensorFlow, PennyLane, Qiskit; circuits simulated)
- Predicted CRISPR gene dependency from gene copy number, using DepMap data
PythonQiskitPennyLaneDeep Learning
external source ↗
Award · 3rd place national hackathon
Oct 2021 — Nov 2021
Recommendation & optimization system awarded 3rd place overall at Ambev’s national hackathon (ABI AcademyHack).
- Modeled consumer profiles and discount elasticity from historical transaction data to maximize LTV and repeat purchases
- Presented at the finalist Demo Day
PythonData ModelingRecommender Systems
external source ↗
career
-
Dec 2025 — Present
Quantitative Systems Developer
Algorithmic trading (MetaTrader 5)
- Proprietary execution and market-monitoring algorithms (Python/MQL5).
- Ran continuously and unattended on a demo account until a full reset in October 2026, with risk-containment layers that act independently of the strategy logic.
-
Mar 2025 — Present
Technology Consultant
Assiny (checkout & payments platform)
- Tracking/pixel scripts with configurable templates served by a dedicated worker; manual conversion logging.
- Modeled webhook ingestion semantics: refunds, chargebacks, fees.
- Market analysis and competitive positioning reports (2026); API integration advisory.
-
Aug 2024 — Present
Tech Lead, Principal Systems Architect & General Manager (Interim CEO)
RoiGenius
- End-to-end technical + executive leadership of a Revenue Attribution & Analytics SaaS: engineering direction, product roadmap, cash flow and GCP cost governance.
- Designed 8 TypeScript/NestJS 11 microservices on Google Cloud Run, backed by Cloud SQL PostgreSQL 15 tuned for high concurrency (max_connections=500).
- Built CalculatorRegistryService standardizing financial metrics across 89 ad/checkout platforms with strict string-based BRL arithmetic — zero rounding drift.
- Engineered async webhook pipelines: instant HTTP 200 ack + Dead Letter Queues with progressive retry windows (1h→16h).
-
Aug 2024 — Present
AI Platform & Distributed Systems Engineer
Independent research & platform engineering
- Engineered a self-hosted distributed compute platform: heterogeneous workers with GPU acceleration for reranking and local inference — fully offline, zero cloud dependency.
- Local-first multi-LLM platform behind a unified gateway with complexity- and cost-based routing.
- Bi-temporal semantic memory: knowledge graph + vector store with valid_from/valid_to versioning — queryable history, not just current state.
- Governance for autonomous agents: 75+ model-agnostic skill modules and 100+ deterministic security hooks.
-
Feb 2023 — Aug 2023
Exchange Student — MSc Computer Science (Spring 2023)
EPITA, Paris · France
- Semester average 15.52/20 (Honors), 20 ECTS validated.
- Grade A highlights: Unix (18.02/20), Python (18.74/20), Blockchain & Bitcoin (20/20 — top of class).
-
Jun 2022 — Aug 2022
Strategic Advisor — Technology Integration
Mamba Culture
- Data ecosystem integration and automation consulting (Intercom, ActiveCampaign, Takeblip).
- Audited and cleaned relational databases; trained teams on data modeling and API governance.
-
Jan 2021 — Jan 2023
Project Manager & Technical Lead
Be Growth Brasil Internet S.A.
- Technical lead (Clube Utua): architectural restructuring that drove 24.5× revenue and user-base growth.
- Led a multidisciplinary R&D squad: rapid prototyping, agile standards reused org-wide.
- Cross-platform mobile/web engineering with Flutter and JavaScript.
-
Sep 2020 — Dec 2020
Teaching Assistant
PUC Minas
- Teaching assistant for Algorithms & Data Structures I and Web Interface Development.
- Supported students in programming logic, asymptotic analysis, core web technologies.
-
2019 — 2025
BSc, Computer Science — GPA 84.16
PUC Minas
- Class President (Mar 2021 — Dec 2022).
- HPC research published at SSCAD 2024 (quantum-classical neural networks).
stack
languages
- TypeScript · Python · C++ · SQL · Bash
cloud & backend
- NestJS 11 · Cloud Run · Cloud SQL · Redis · Docker · Firebase
distributed AI
- local-first inference · calibrated decision layers · agentic orchestration
frontend & edge
- React 18 · Vite · CoreUI Pro · Cloudflare Workers
quant
- MetaTrader 5 (MQL5/Python)