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
CarnetClair AI Systems helps organizations design, evaluate, and implement reliable GenAI, agentic AI, and applied machine learning solutions.
Every layer is a design decision, not a black box.
Capabilities
Focused support for the parts of AI delivery where technical depth, system judgment, and measurable reliability matter most.
01
Retrieval architecture, chunking, hybrid search, reranking, grounding, citation quality, and end-to-end evaluation.
02
Stateful workflows, tool use, deterministic controls, bounded reasoning, approval gates, and failure handling.
03
Test design, component and trajectory evaluation, adversarial cases, observability, and human oversight.
04
Technical discovery, solution architecture, focused prototypes, design reviews, and production-readiness planning.
05
Serving architecture, latency and throughput analysis, caching, batching, GPU utilization, and cost tradeoffs.
The CarnetClair difference
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.
Use frameworks deliberately, with clear knowledge of the behavior they encapsulate.
Measure components and end-to-end outcomes rather than relying on a polished demonstration.
Account for reliability, latency, cost, failure modes, and human control from the beginning.
About CarnetClair
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
Public projects that examine orchestration, model architecture, and inference performance from first principles through measurable implementation.
A bounded, checkpointed workflow with hybrid retrieval, structured outputs, human approval, resumable state, and reproducible evaluation.
View repositoryA GPT-style implementation covering tokenization, attention, transformer blocks, pretraining, generation, classification, and instruction fine-tuning.
View repositoryReproducible single-GPU experiments measuring concurrency, prefill, decode, prefix caching, throughput, and latency on an NVIDIA L40S.
View repositoryGlobal collaboration
CarnetClair supports remote engagements with organizations across the United States and internationally.
Contact us
For project inquiries, architecture reviews, technical advisory work, or focused AI system development.
Prefer email?
info@carnetclair.com