Amazon Nova Pro

Cost & access

Amazon Nova Pro is proprietary via Amazon. Typical billing: input + output tokens; ChatGPT-style subscriptions are separate from API access. With a 300K tokens context window, long PDFs or chat histories increase input tokens quickly—trim history or summarize older turns in production.

Functional understanding

• AWS Nova family model for multimodal enterprise workloads on Bedrock.
• Modalities: text, image, video · License: Proprietary · Released 2024-12-03.
• Best-fit workflows for this model:
• • Document OCR, chart/diagram understanding, and visual QA over screenshots or PDFs.
• • Short video understanding, scene description, and frame-level analysis.

Technical foundation

• Amazon reports Undisclosed parameters; training data: Undisclosed.
• Context: 300K tokens. Open weights: no.
• Amazon Nova Pro is positioned as a vision model in the Amazon lineup.

First API call

Follow Amazon's official SDK for Amazon Nova Pro; use model id "amazon-nova-pro" from their docs.

# See https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html
# Model id: amazon-nova-pro

Important technical topics

• Prompting Amazon Nova Pro: be explicit about output format. Weak: "Analyze this." Better: "Return JSON with fields id, total, date for Amazon billing data."
• Temperature: use 0–0.3 for extraction and compliance on Amazon Nova Pro; 0.7–1.0 for brainstorming.
• Tokens: Amazon Nova Pro bills by tokens (~¾ word each). Undisclosed parameters affect capability; your bill is driven by context length and call volume.
• Context window (300K tokens): everything in one request—system prompt, tools, RAG chunks, and history—must fit. Truncate or summarize when approaching the limit for Amazon Nova Pro.
• Vision tokens: images in Amazon Nova Pro consume extra tokens (often tiled patches)—compress resolution when cost matters.

Real enterprise patterns

• Pipeline: OCR/layout → Amazon Nova Pro for field extraction → rules engine for validation.
• Store original images; log model version per request for audit.
• Redact PII in images before sending to third-party APIs unless self-hosting.
• Fallback to smaller vision model for simple yes/no checks.

Production & security

• Secrets: never commit keys for Amazon Nova Pro; use vault + per-environment rotation.
• PII: mask before inference; log redacted prompts only.
• Observability: trace id per request; log model=amazon-nova-pro, tokens in/out, latency.
• Rate limits: handle Amazon 429/5xx with exponential backoff and circuit breakers.
• Guardrails: schema-validate JSON; block disallowed topics; cross-check numbers against source docs.

Mini projects with this model

• Invoice OCR: Amazon Nova Pro extracts line items → CSV.
• UI regression: compare screenshots, describe visual diffs.
• Safety checklist: verify PPE in warehouse photos.
• Catalog enrichment: generate alt text from product images.

Suggested stack

• Language: Python 3.11+
• LLM: Amazon Nova Pro (Amazon official SDK)
• UI: Streamlit or Next.js for internal tools
• APIs: FastAPI
• Vector DB (RAG): Pinecone / Chroma / pgvector
• OCR helper: Azure Document Intelligence or Tesseract pre-pass

Learning path

• Python basics
• HTTP/REST and environment variables
• Amazon authentication and Amazon Nova Pro model id (amazon-nova-pro)
• First successful call to Amazon Nova Pro
• Prompt design and JSON / structured outputs
• Image encoding, resolution, and token costs
• RAG
• Tool use / function calling
• Evals and regression sets
• Production deploy + monitoring