Gen AI and MLOps

Domino Data Lab

Domino gives data science and AI teams one governed workspace for building, deploying and monitoring models and agents, with tracing and evaluation built in, across cloud and on-premises infrastructure. We set up projects, environments and compute, automate release pipelines for models and agents, and put approvals, audit trails and cost controls in place, so regulated teams can move quickly and still show their work.

Spiral

Spiral is a database for multimodal AI data, built on the open-source Vortex columnar format and running natively on object storage such as Amazon S3 and Google Cloud Storage. It handles video, audio, images, traces and tabular data, keeps datasets versioned, and loads them straight into GPU memory for PyTorch training. We use it to turn raw multimodal data into governed, tested datasets for training and evaluation.

Denodo

The Denodo Platform creates a logical data layer across warehouses, data lakes, SAP and operational systems without copying the data. Its AI SDK gives LLMs and agents natural-language access to governed, real-time enterprise data, with a Model Context Protocol (MCP) server and DeepQuery for multi-step research questions. We design the semantic layer, security policies and query paths that keep answers accurate, current and traceable.

Platforms we build on

Three platforms we use to take Gen AI and MLOps into production.

Gen AI teams

For engineers building LLM applications. We help you choose and evaluate models, design prompts and retrieval, add guardrails, and measure quality, latency and cost with repeatable test sets, so a promising demo becomes a service you can support.

MLOps teams

For the people who keep models running. We build CI/CD for models, prompts and agents, with registries, automated evaluation gates, drift and cost monitoring, and rollback. LLMOps then sits on the same disciplined footing as your classical machine learning.

Data teams

For data engineers and data owners. We make enterprise data ready for AI: pipelines, quality checks, metadata and lineage, access policies, vector and graph indexes, and a semantic layer that gives every model the same trusted definitions.

Quant teams

For quantitative analysts and researchers. We provide reproducible research environments, scalable compute for backtesting and simulation, and governed paths from notebook to production, with model documentation and validation evidence ready for model-risk review.

Who we work with

Four teams, each with its own priorities and its own definition of done.

RAG

Retrieval-augmented generation grounds an LLM’s answers in your own documents and data. We build the ingestion, chunking, embedding and retrieval pipeline, add citations and access controls, and test answer quality against real questions before users see it.

Graph RAG

Graph RAG adds a knowledge graph to retrieval, so the model can follow relationships between customers, products, contracts and events. It suits questions that span many documents. We model the graph, populate it from your sources, and combine graph and vector search.

Agentic AI

Agents plan a task, call tools and APIs, and check their own results. We define each agent’s scope, tools and permissions, add evaluation and tracing, and keep a person in the loop for decisions that carry risk.

Agentic workflows

Agentic workflows coordinate several agents and human approvals across an end-to-end process such as onboarding, reconciliation or reporting. We design the orchestration, hand-offs, exception handling and audit trail, then pilot with one team before scaling.

Use cases we deliver

Four patterns that cover most enterprise Gen AI work today.

Plan your first Gen AI use case

Tell us which team and use case you want to start with. We’ll reply within one business day.