Ingest Labs gives martech customers server-side tagging, a Media Data Platform, and GenAI products like AdShield. The one capability its customers had to leave the platform for was ad creative. Avahi built a production-grade GenAI pipeline on AWS that turns a plain-English prompt into a finished, compliant, on-brand ad, so customers can create and launch campaigns without ever leaving Ingest Labs.
Amazon Bedrock · Nova Canvas · Claude
From a plain-English prompt to a finished, compliant creative, fully automated
Nova Canvas, Claude Sonnet & Claude Vision, deployed by Avahi
Ingest Labs provides martech solutions across a broad space. Specifically, it runs a server-side tagging platform, its own Media Data Platform (MDP), and, more recently, GenAI-native products such as AdShield.
Its customers almost all run ads, and every ad needs creative. In the company’s words: “Our customers predominantly all run ads, and when they are running ads, they need creatives.”
Ad creative was the one part of that workflow Ingest Labs did not yet offer inside its own ecosystem, which set up the opportunity Avahi was brought in to build.
Advertising Technology (Martech)
Claude Sonnet & Claude Vision · Amazon Nova Canvas
Avinash Gujjar, VP Alliances & Partnerships
Amazon Bedrock · S3 · ECS Fargate · Application Load Balancer
Avahi, AWS Premier Tier Services Partner
Because Ingest Labs did not offer creative generation, customers had to produce ad creative somewhere else and bring it back. As the company put it: “The customers had to go and fish for it outside of our platform.” Closing that gap meant meeting three requirements at once:
Ingest Labs did not start with a partner in mind. AWS brought Avahi into the conversation, introducing it as an AWS-certified SI partner suited to the work.
Avahi built a pipeline where a user gives a prompt in plain English and gets back a finished ad creative. As the company described it: “Using simple English, I should be able to say, create an ad creative with a person sitting on stairs holding a guitar, and in the backdrop let there be a sunset.” It sounds simple, but, as they noted, there are a lot of complexities involved: every result has to stay on-brand and legally compliant. The pipeline handles it in six moves:
A four-layer guardrail framework screens every prompt: deterministic rules flag explicit violations instantly, then Claude Sonnet on Amazon Bedrock adds semantic classification to catch nuanced, context-dependent risks rules alone would miss.
Prompts that pass review are normalized and enhanced to get the best possible image-generation quality before any pixels are made.
A unified service-abstraction layer routes generation to Amazon Bedrock’s Nova Canvas or Google Gemini, treating both interchangeably: the strengths of multiple engines, with no brittle model-specific code.
Each generated image is checked by Claude Vision for visual-level safety that text guardrails cannot see. Anything that fails triggers automatic regeneration, so only compliant creatives advance.
A configuration-driven engine built with Pillow applies brand elements, layout rules, and platform-specific formatting consistently, so every creative stays aligned with the customer’s brand story.
An invisible steganographic watermark adds cryptographic provenance, and finished assets land in Amazon S3. LangGraph orchestrates the whole graph with fail-open and retry logic, containerized with Docker on AWS ECS Fargate behind an Application Load Balancer.
Customers create ad creative and launch campaigns without leaving Ingest Labs.
A simple text prompt produces a finished, on-brand, legally compliant creative.
A new capability Ingest Labs can take to existing customers and new prospects.
The company describes the impact as two-fold. First, Ingest Labs can market a brand-new capability to customers and prospects, widening its reach. Second, existing customers can create ad creatives and launch campaigns from within the platform, without ever having to worry about leaving it.
Without leaving my platform, I can create those ads and launch campaigns, all from one place. That is the business impact this created.
Because the pipeline is model-agnostic by design, new generation or classification models can drop in without reworking the system, so Ingest Labs keeps pace as the technology landscape shifts.
The same architecture is built to scale across Ingest Labs’ market: advertising agencies, e-commerce brands, social-media marketing teams, and content-generation platforms.
Swap or add Bedrock and third-party models behind the service-abstraction layer, with no pipeline rewrite.
A simple text prompt produces a finished, on-brand, legally compliant creative.
Steganographic watermarking gives every asset cryptographic traceability for regulated advertising.
Avahi is an AWS Premier Tier Services Partner. Our engineers build cloud migrations and generative AI workloads directly inside your AWS account, the same way we built Ingest Labs’ prompt-to-placement pipeline. Bring your goals; we’ll scope the path.
AWS-funded proof of concept · Senior engineers · Your cloud, your data
