GenAI Workflows

Overview of how Valohai enables reproducible, scalable, and integrable GenAI workflows.

Valohai provides the foundation for running, comparing, and governing GenAI models and pipelines at scale.

Why Valohai for GenAI

Generative AI workflows aren’t just prompt engineering or model calls. They involve data pipelines, evaluations, retraining loops, and governance — the same challenges as traditional ML, but with new complexity: non-deterministic outputs, subjective metrics, and constantly evolving datasets.

Valohai brings engineering discipline to that process:

• Reproducibility: Every model, dataset, and evaluation run is versioned automatically.

• Pipelines, not notebooks: Move from one-off experiments to governed workflows.

• Human approval gates: Add subjective or compliance reviews into automated pipelines.

• Model Catalog: Compare versions side-by-side with metrics, datasets, and lineage in one place.

What You Can Do

Workflow	Description	Learn More
Evaluate GenAI applications	Run automated and human-in-the-loop validations using versioned datasets and metrics.	Evaluating and Validating GenAI Applications
Retrain and promote safely	Automate retraining pipelines with regression checks and approval gates.	Retraining and Updating GenAI Models
Finetune large models	Adapt base LLMs to your domain or tone using reproducible datasets and tracked experiments.	Finetuning LLMs in Valohai
Evaluate multiple models & providers	Run a provider‑agnostic leaderboard (OpenAI, Anthropic, Llama) on a fixed evaluation dataset; track quality, latency, and cost.	Evaluating Multiple Models and Providers

Integrations and Extensibility

Valohai integrates smoothly with the tools and services you already use:

• Webhooks: Trigger external systems or notifications when executions or datasets complete.

• Notifications: Connect to Slack, Teams, or your incident tooling for instant updates.

• REST API: Every operation in the UI, executions, datasets, model management, is also available through the Valohai API. Build custom dashboards, CI/CD hooks, or integrate with internal GenAI platforms directly.

• Flexible compute: Run workloads on cloud, hybrid, or on-prem environments using your own GPUs or clusters.

Valohai doesn’t replace your GenAI models it gives you the reproducible environment to train, evaluate, and deploy them safely and repeatedly.

Typical Use Cases

• LLM evaluation and comparison (track BLEU, ROUGE, or human ratings)

• RAG pipeline retraining (version embeddings, corpora, and retrievers)

• Domain finetuning (adapting base models to enterprise data)

• Governed deployment (approval gates before production rollout)

Learn More

• Valohai Pipelines

• Valohai Model Catalog

• Automation with Webhooks & Notifications

• RAG Example