Senior Software Engineer · Technical Lead · AI/ML Engineer📍 Florianópolis, Brazil

Pedro
Soares

I build software products, distributed systems, and AI infrastructure — from SaaS platforms and cloud architectures to language models and GPU systems.

12+
YEARS ENGINEERING
6
STACK LAYERS
~100M
DB INSERTS/DAY
~60M
LLM PARAMS TRAINED
1st
SALESFORCE SDK IN FLUTTER · BRAZIL
3+
SAAS PRODUCTS SHIPPED
About

Engineering with
product depth.

Software Engineer with 12+ years of experience designing, building, and scaling products across web, mobile, cloud, AI, real-time systems, and game development. Strong background in software architecture, distributed systems, serverless computing, and technical leadership.

Combines deep technical expertise with a strong product mindset — focused on measurable outcomes: performance, reliability, scalability, and cost efficiency.

Consulting & Advisory

Through Carbonis Fundatus, I help companies overcome technical bottlenecks and scale their platforms.

→Re-architecting legacy systems and designing cloud-native architectures
→Scaling platforms to support increased traffic and demand
→Reducing cloud and infrastructure costs (FinOps)
→Modernizing development processes and engineering workflows
→Mentoring development teams and technical leaders
→Defining technology strategies aligned with business objectives
Industries
AIEdTechMarketing AutomationE-CommerceCustomer ServiceSaaSGamingHealth & InsuranceReal-Time Systems
Work by Domain

How my work connects.

Hover a domain to see the projects that live there. Most span multiple layers.

SOFTWARE
├
├
├
└
Product Engineering
SaaS · CRM · AI-powered products · API design
CRM & Marketing Automation
Visual workflow builder, Email/SMS/WhatsApp automation
Emilia — AI Customer Service
Virtual assistant, decision trees, education vertical
Marketing Automation Engine
Graph-based execution, state machines, scheduling
AI Product Infrastructure
Model integration, conversation APIs, backend
Career

Experience

Tarmac.IO
Senior Software Engineer
Contract · Remote · Brazil
PostgreSQLTerraformRemote
Oct 2024 – Present
Carbonis Fundatus
Founder
Self-employed · Remote · Florianópolis, Brazil

Founded and built CRM & marketing automation platform, AI customer service product (Emilia), and conducted AI/ML research including LLM training from scratch and knowledge distillation experiments.

PythonRustAI/MLSaaS
Oct 2023 – Present
NE Digital · Grupo JCPM
Senior Tech Lead / Software Engineer
Recife, Brazil

Led the technical development of software products, working across architecture, backend, mobile, cloud infrastructure, and distributed systems. Delivered the first native Salesforce SDK integration for Flutter in Brazil — fully featured on both iOS and Android — in two weeks leading a two-developer team.

Tech LeadFlutterSalesforce SDKNode.jsRustAWS
2023 – 2024
Vai de Bolsa
Senior Software Engineer
Brazil

Designed and developed a new e-commerce platform from the ground up, migrating the company's existing WordPress-based e-commerce to a custom solution. Responsible for architecture, backend, frontend, integrations, and migration strategy.

E-commerceArchitectureBackendFrontend
2022 – 2023
Work

Selected Projects

Click any card to expand.

01AI / ML

LLM From Scratch

Training a ~60M parameter language model end-to-end

PyTorchTransformersCUDAHugging FacePythonDataset Curation
Expand ↓
02Research

Continuum Processor

Experimental language-model architecture and training research

PyTorchArchitecture ResearchTokenizersData ScalingEvaluation
Expand ↓
03Systems

High-Scale Facial Recognition

Computer vision pipeline for sports photography at 100M inserts/day

RustAWS RekognitionLambda ARM64SQSFargateAurora ServerlessPostgreSQL
Expand ↓
04Product

CRM & Marketing Automation

Composable SaaS CRM with a visual event-driven workflow engine

Node.jsVue 3Nuxt 3PostgreSQLEvent-DrivenSaaSWhatsAppSMS
Expand ↓
05AI / ML

Emilia — AI Customer Service

AI-powered virtual assistant for automated customer support

LLMConversational AIDecision TreesPythonSaaSEducation
Expand ↓
06Research

Knowledge Distillation Research

Dataset-free training via teacher-to-student knowledge transfer

Knowledge DistillationSynthetic DataTeacher/StudentPyTorchLLM Research
Expand ↓
07Systems

Rust Image Processing

High-performance image pipeline with GPU/CPU optimization

RustONNX RuntimeFacial DetectionEmbeddingsGPU
Expand ↓
08Cloud

Serverless High-Throughput Processing

AWS serverless architecture for high-volume async data ingestion

AWS LambdaSQSFargateARM64ServerlessCost Optimization
Expand ↓
09Systems

Marketing Automation Engine

Graph-based workflow execution engine with state management

Graph ExecutionState MachinesEvent-DrivenSchedulingNode.jsPostgreSQL
Expand ↓
10Product

AI Product Infrastructure

Backend infrastructure for production AI-powered products

PythonFastAPILLM IntegrationConversation APIBackend
Expand ↓
11Systems

Whisper Voice Recognition → Godot

Porting OpenAI Whisper to Godot via Rust with streaming and Opus

RustWhisperGodotOpus CodecStreaming AudioGDExtension
Expand ↓
Ongoing

Research

Active experiments in AI training, model architecture, and knowledge transfer.

KNOWLEDGE DISTILLATION

"Can a teacher model's knowledge train a student without any traditional dataset?"

Investigating synthetic data generation via teacher models as a path to dataset-free pretraining. Early signals: architectural alignment between teacher and student matters as much as data volume.

CONTINUUM PROCESSOR

"Is architectural failure often a data failure in disguise?"

A custom LM architecture that showed dramatically improved generation with more diverse training data. The key finding: the bottleneck was corpus diversity, not the architecture itself.

Contact

Let's build
something deep.

Open to Senior / Staff / Principal Engineer, Technical Lead, Software Architect, and Founding Engineer roles. Available for consulting on AI systems, distributed architecture, and cloud infrastructure.

contato@pedrosoares.netLinkedInGitHub