01
GenAI & RAG systems
Retrieval architecture, chunking, hybrid search, reranking, grounding, citation quality, and end-to-end evaluation.
- Architecture and implementation
- Retrieval quality diagnostics
- Grounded generation
Applied AI consulting · Remote worldwide
Production AI systems,
built to be understood.
CarnetClair AI Systems helps organizations design, evaluate, and implement reliable GenAI, agentic AI, and applied machine learning solutions.
System view
Production-minded
05
Product layerUseful, clear, measurable
04
Evaluation & safeguardsQuality, reliability, human oversight
03
OrchestrationAgents, tools, state, bounded workflows
02
Retrieval & contextGrounding, ranking, evidence
01
Models & inferencePerformance, serving, cost
Every layer is a design decision, not a black box.
Ph.D.Applied Mathematics
13+ yearsPost-Ph.D. experience
Microsoft · Intel · Google Cloud Premier PartnerEnterprise experience
EN · FR · ES · PTInternational collaboration
Capabilities
From first architecture decision to production behavior.
Focused support for the parts of AI delivery where technical depth, system judgment, and measurable reliability matter most.
02
Agentic AI systems
Stateful workflows, tool use, deterministic controls, bounded reasoning, approval gates, and failure handling.
03
Evaluation & reliability
Test design, component and trajectory evaluation, adversarial cases, observability, and human oversight.
04
Architecture & prototyping
Technical discovery, solution architecture, focused prototypes, design reviews, and production-readiness planning.
05
LLM inference & cost
Serving architecture, latency and throughput analysis, caching, batching, GPU utilization, and cost tradeoffs.
The CarnetClair difference
Clarity all the way down.
CarnetClair combines application-level AI engineering with a practical understanding of what is happening beneath the interface.
Models, retrieval, orchestration, evaluation, serving, performance, and cost are treated as one connected system.
Understand before abstracting
Use frameworks deliberately, with clear knowledge of the behavior they encapsulate.
Evaluate the full system
Measure components and end-to-end outcomes rather than relying on a polished demonstration.
Design for production constraints
Account for reliability, latency, cost, failure modes, and human control from the beginning.
About CarnetClair
Technical leadership grounded in mathematics and delivery.
CarnetClair AI Systems is an applied AI consultancy helping organizations turn complex GenAI and machine learning ideas into reliable, understandable production systems.
CarnetClair AI Systems was founded by Dr. Delphine Mico Umutesi, a Principal Applied AI Engineer and Data & Applied Scientist with a Ph.D. in Mathematics and more than 13 years of post-Ph.D. experience. She has delivered production AI, machine learning, agentic AI, and optimization systems across enterprise environments, including Microsoft, Intel, and a Google Cloud Premier Partner.
Her work spans production RAG, agentic workflows, deep learning, optimization, evaluation, and GPU-aware LLM inference. The common thread is translating technical depth into systems that are useful, explainable, and engineered to perform.
Selected technical work
Depth demonstrated in working systems.
Public projects that examine orchestration, model architecture, and inference performance from first principles through measurable implementation.
Production-oriented retention agent
A bounded, checkpointed workflow with hybrid retrieval, structured outputs, human approval, resumable state, and reproducible evaluation.
View repositoryBuilding an LLM from scratch
A GPT-style implementation covering tokenization, attention, transformer blocks, pretraining, generation, classification, and instruction fine-tuning.
View repositoryLLM inference performance
Reproducible single-GPU experiments measuring concurrency, prefill, decode, prefix caching, throughput, and latency on an NVIDIA L40S.
View repositoryGlobal collaboration
Based in Miami.
Built to work across borders.
CarnetClair supports remote engagements with organizations across the United States and internationally.
English Fluent
French Fluent
Spanish Professional collaboration
Portuguese Professional collaboration
Contact us
Bring the problem.
We will make the system clear.
For project inquiries, architecture reviews, technical advisory work, or focused AI system development.
Prefer email?
info@carnetclair.com