Luciana
Reynaud Ferreira
LLM Production Systems Engineer
FinOps · Reliability · Governance
Brazil · Remote

I design AI systems at the point where prototypes become infrastructure: when model-driven workflows need to be observable, economically justified, operationally reliable, and safe to scale. My work sits at the intersection of serving architecture, model routing, telemetry, cost attribution, and governance — with cost, reliability, compliance, and auditability treated as engineering constraints from the start. I am most useful when the problem is no longer "can we use AI?" but "can we operate it repeatedly, explain its behavior, and make its economics hold under real usage?"

// Selected Impact
50×
Cost reduction through operating-model redesign and model routing — projected spend reduced from roughly $210 to $4.20/month, with 3.5% budget utilization and documented headroom for source expansion.
Hybrid ranking
Built an auditable relevance classifier combining embeddings, kNN retrieval, and LLM scoring on top of a 280-example labeled reference set — producing rankings that are inspectable, correctable, and improvable without retraining or fine-tuning.
1.07M
Users reached in a single WABA broadcast for Recovery at Blip, generating approximately R$2.3M in closed-deal revenue under regulated operating constraints.
Cost model
$0.14/day baseline documented, 96.5% budget headroom quantified, 6× source expansion modeled — translating architecture into executive financial decision.
// Focus
LLM FinOps & Cost Attribution
Token budget design, model routing, batch vs. on-demand cost analysis, and spend visibility across production AI systems.
Observability & Telemetry
Structured logging, per-source instrumentation, tracing, and operational visibility for model-driven systems in production — with OpenTelemetry as the target baseline.
Production Reliability
Deployment architecture, evaluation loops, health surfaces, and monitoring workflows that detect and correct behavioral drift.
AI Governance & Compliance
Risk-aware system design, decision logging, LGPD-aware architecture, and auditability for regulated or high-consequence environments.
// Experience
Feb 2025 – Present
Remote · Brazil
AI Systems Engineer
MakeOne Lab
  • Built production observability layer for the lab's LLM stack: per-source structured logging across 52 ingestion endpoints, per-execution ranker run logs with classification tracing, health endpoints, and cost attribution at the token and model level — establishing the instrumentation baseline for future OpenTelemetry integration.
  • RondaPress — architecture and product intelligence: Auditable relevance ranking across 52 sources and 150–300 articles/day — each decision traceable to its nearest labeled neighbors, enabling editorial correction without retraining. Built on a hybrid kNN + LLM pipeline anchored in a 280-example labeled dataset.
  • RondaPress — serving model and economics: Reduced projected monthly AI spend from roughly $210 to $4.20 — a 50× reduction — by replacing per-user on-demand inference with centralized scheduled processing plus instant dashboard access. Implemented model routing with 99% of classification traffic on GPT-4o-mini, maintaining 3.5% budget utilization with clear headroom for expansion.
  • RondaPress — operational visibility: System is debuggable, auditable, and economically predictable rather than prompt-driven and opaque — health endpoints, ranked JSON exports, scraper logs, and per-source execution visibility built in from the start.
  • OneStart Sales Intelligence: Built an end-to-end pipeline for transcription, structured extraction, stakeholder and pain-point mapping, enrichment from LinkedIn and company websites, and automated strategic sales report generation. Implemented FastAPI backend, Postgres storage, and dual deployment across Docker and serverless surfaces.
  • Contributed to Fraport airport intelligence on Databricks, supporting real-time dashboards for passenger presence and connectivity behavior across Fortaleza and Porto Alegre.
Feb 2024 – Feb 2025
Ribeirão Preto · Brazil
Technical Lead
Scalar School · Human Rights Foundation Grant
  • Secured and managed USD 25,000 from the Human Rights Foundation as sole technical lead, designing the full program architecture, curriculum, and technical content for distributed systems and open financial infrastructure.
  • Scaled to 150+ developers across university chapters, delivering protocol-level technical education and establishing structured community formation around systems reasoning.
Jun 2023 – Mar 2024
Remote · Brazil
AI Systems Engineer
Voz AI · independent contractor
  • Embedded LGPD, Central Bank regulations, and KYC requirements as design constraints from the first flow, treating compliance as architecture rather than review-layer overhead.
  • Designed and specified the conversational AI system for BMG Bank: 20 modular Dialogflow CX flows with full error handling, unsupported-format exits, session parameter architecture, and behavioral event tracking defined alongside the data engineering layer.
Nov 2021 – Jun 2023
Remote · Brazil
AI Systems Engineer
Blip
  • Owned production customer-facing AI systems for Itaú Personalité, Itaú Credimob, Bradesco Alelo, Recovery, and HDI Seguros across regulated financial and insurance environments.
  • Designed an anti-fraud WABA broadcast to 1,077,000 users for Recovery, generating approximately R$2.3M in closed-deal revenue within LGPD and WhatsApp Business API constraints.
  • Ran iterative model-quality cycles from raw production logs, identifying fallback concentration, low-confidence behavior, and audience-specific vocabulary to inform retraining and system refinement.
  • Designed in-bot NPS, CSAT, and CES measurement flows on WhatsApp and Instagram, instrumenting live systems for continuous quality signal.
// Stack
Core Specialty
LLM FinOps Cost Attribution Model Routing Observability OpenTelemetry AI Governance
Infrastructure & Production
Python FastAPI Docker Linux · Ubuntu Server GitHub Actions Cloudflare Tunnels Postgres Databricks Distributed Systems
AI & ML
OpenAI API Anthropic API Whisper Hugging Face Vector Search Embeddings RAG Pipelines Pydantic
Compliance & Governance
LGPD KYC Architecture Drift Detection Decision Logging Evaluation Loops Audit Trails
// Education
2024
B.Tech in Systems Analysis & Development
São Paulo State College of Technology (FATEC-SP)
Discrete Mathematics · Calculus · Applied Statistics · Computer Architecture · Algorithms & Data Structures · Software Engineering
// Contact
github linkedin
Open to remote