Choose a model
GLiNER: structured decisions for AI systems
Agent and browser workflows
- Browser use and automation: identify page intent, relevant fields, warnings, and actions that need confirmation.
- Context compaction: extract decisions, tasks, deadlines, and user preferences from conversations or agent traces.
- Model routing: classify domain, complexity, latency requirements, and risk before selecting a model or workflow.
- Agent supervision: detect drift, repeated failures, missing deliverables, risky tool calls, irrelevant context, poor output quality, or when a human should step in.
Retrieval, data, and research
- RAG: select relevant passages before generation and check retrieved context for relevance or conflicts.
- Knowledge graphs and agent memory: extract entities and relationships from documents or conversations into a queryable fact layer.
- Text classification: sort emails, tickets, and documents into operational labels such as spam, urgent, refund request, or sales inquiry.
- PII detection and redaction: identify sensitive data before storage, indexing, or model calls.
- Forecasting and market intelligence: turn customer feedback, filings, transcripts, and news into signals such as demand, supply concerns, competitor mentions, and management changes.
Safety and quality checks
- LLM guardrails and moderation: screen prompts and outputs for harmful content, sensitive data, and policy violations.
- Prompt-injection detection: flag attempts to override instructions, exfiltrate data, or manipulate tool use.
- Semantic code linting: check generated code and pull requests for conventions that syntax alone cannot catch.
Fastino-Nemotron-3.5-Lightning-Healthcare
- Clinical documentation: summarize notes, encounters, patient timelines, and discharge documentation.
- Chart preparation: surface relevant diagnoses, medications, symptoms, and recent events before an encounter.
- Patient communication: draft plain-language, clinician-approved care instructions and responses.
- Clinical research: organize evidence, summarize literature, and support trial-matching review.
- Documentation review: flag missing sections, conflicting dates, and unclear follow-up items.
Fastino-Nemotron-3.5-Lightning-Finance
- Filing and annual-report Q&A: answer research questions over reports, filings, and dense tables.
- Financial document summaries: produce structured notes from earnings releases, investor presentations, and reports.
- Table and metric analysis: locate relevant figures, calculate comparisons, and show the formula used.
- Multi-turn financial research: continue a line of inquiry while keeping the relevant sources and assumptions visible.
- Earnings review and forecasting inputs: extract reported metrics, guidance, narrative signals, and market changes for analyst or forecasting workflows.
- Compliance-aware research: pair generation with GLiNER checks for PII, required disclosures, sensitive claims, and review routing.
Combine models in one workflow
Start with a measurable task and a small evaluation set built from permissioned examples. For GLiNER, define the labels or schema. For a Lightning model, define the source documents, required checks, and review step.