From concept to a working, demonstrable prototype
Human-labeled reactions used to validate the sentiment scoring
Scores delivered without a written rationale from the judging models
Untamed Planet
New York, USA
Media Production / Sustainability Communications Consulting
: Amazon Bedrock, AWS Lambda, Amazon API Gateway, AWS Step Functions, Amazon S3, Amazon DynamoDB, Amazon CloudFront, AWS IAM, Amazon CloudWatch, AWS STS
Untamed Planet is a production company that has spent roughly a decade creating documentaries and factual programming, and is expanding into storytelling consulting to help organizations communicate sustainability and impact more effectively. Its Narrative Lab initiative set out to move sustainability messaging advice from individual expertise toward a data-informed method. Avahi built a working prototype that ingests public content about a sustainability campaign and returns a structured, evidence-based view of how that campaign is landing, in a four-week engagement, giving Untamed Planet a foundation to validate the concept with prospective clients.
Untamed Planet is a New York based production company with a history of documentaries and factual programming for major broadcasters and media organizations. Its Narrative Lab initiative extends that work into storytelling consulting, helping organizations ground sustainability and impact messaging in evidence rather than individual judgment alone.
Untamed Planet’s sustainability messaging consulting relied on the subjective expertise of individual consultants, delivered through multi-day, in-person workshops and roadmaps. Message effectiveness was judged manually, without a repeatable method, and insights weren’t systematically benchmarked against competitors or across markets. As the Narrative Lab initiative looked to scale that advisory work, the team needed a way to ground its recommendations in measurable, evidence-based signals rather than expert judgment alone.
The prototype needed to run multiple foundation models against retrieved content to reduce single-model bias in its scoring, and to do so on infrastructure that could move from prototype to a production build without a redesign. Amazon Bedrock gave Avahi access to several foundation models for an LLM-as-judge scoring approach, while a serverless architecture built on Amazon API Gateway, AWS Lambda, AWS Step Functions, Amazon S3, and Amazon DynamoDB let the prototype ingest and process content asynchronously, fanning ingestion out in parallel, without managing dedicated servers or paying for idle compute between runs. Amazon CloudFront serves the reporting dashboard, and AWS IAM and CloudWatch give the team least-privilege access control and a full evidence trail from a single log group per function.
As a premier-tier AWS partner with experience applying generative AI to real analytical workflows, Avahi could move from concept to a working, demonstrable prototype within a defined engagement window. Untamed Planet needed a partner who could design a scoring methodology from scratch, validate it against a labeled reference sample, and hand over a foundation the team could build on for a production version, rather than a one-off proof of a single idea.
Avahi designed and built an end-to-end prototype that takes a submitted campaign, a URL or keyword paired with an organization, and ingests content about it from a confirmed set of public sources through several parallel ingestion adapters. The pipeline captures publicly visible reach and engagement signals where a source exposes them, and falls back to content-based signals, such as volume of coverage and number of distinct sources, where it doesn’t, so every report is clear about which figures are directly measured and which are inferred.
Retrieved reactions are classified into a defined set of sentiment and tone categories, refined over the course of the engagement to sharpen how skepticism is distinguished from ordinary criticism, and the campaign is benchmarked against a comparable organization on the same signals. Each report generates an effectiveness and credibility score using an LLM-as-judge approach across multiple foundation models on Amazon Bedrock, with each model’s assessment paired with a written rationale rather than a bare number, so the reasoning behind each score is visible and reviewable.
A single-page interface lets a user submit a campaign, trigger the analysis pipeline, and review the generated report end to end, with every figure in the report traceable back to the source item behind it. Avahi validated the sentiment approach against a client-provided labeled reference sample spanning several hundred human-labeled reactions, refined the classification categories based on review feedback, and handed over the full source code, configuration, deployment automation, and architecture documentation.
Untamed Planet came out of this engagement with a working, demonstrable prototype that grounds sustainability messaging assessments in retrieved public evidence rather than individual expert judgment alone. The team validated the approach end to end against sample campaigns, reviewed the generated reports, and now has a documented foundation, source code and deployment automation included, to pursue a production build and validate the concept with prospective clients.
Untamed Planet
New York, USA
Media Production / Sustainability Communications Consulting
: Amazon Bedrock, AWS Lambda, Amazon API Gateway, AWS Step Functions, Amazon S3, Amazon DynamoDB, Amazon CloudFront, AWS IAM, Amazon CloudWatch, AWS STS
Untamed Planet is a production company that has spent roughly a decade creating documentaries and factual programming, and is expanding into storytelling consulting to help organizations communicate sustainability and impact more effectively. Its Narrative Lab initiative set out to move sustainability messaging advice from individual expertise toward a data-informed method. Avahi built a working prototype that ingests public content about a sustainability campaign and returns a structured, evidence-based view of how that campaign is landing, in a four-week engagement, giving Untamed Planet a foundation to validate the concept with prospective clients.
Untamed Planet is a New York based production company with a history of documentaries and factual programming for major broadcasters and media organizations. Its Narrative Lab initiative extends that work into storytelling consulting, helping organizations ground sustainability and impact messaging in evidence rather than individual judgment alone.
Untamed Planet’s sustainability messaging consulting relied on the subjective expertise of individual consultants, delivered through multi-day, in-person workshops and roadmaps. Message effectiveness was judged manually, without a repeatable method, and insights weren’t systematically benchmarked against competitors or across markets. As the Narrative Lab initiative looked to scale that advisory work, the team needed a way to ground its recommendations in measurable, evidence-based signals rather than expert judgment alone.
The prototype needed to run multiple foundation models against retrieved content to reduce single-model bias in its scoring, and to do so on infrastructure that could move from prototype to a production build without a redesign. Amazon Bedrock gave Avahi access to several foundation models for an LLM-as-judge scoring approach, while a serverless architecture built on Amazon API Gateway, AWS Lambda, AWS Step Functions, Amazon S3, and Amazon DynamoDB let the prototype ingest and process content asynchronously, fanning ingestion out in parallel, without managing dedicated servers or paying for idle compute between runs. Amazon CloudFront serves the reporting dashboard, and AWS IAM and CloudWatch give the team least-privilege access control and a full evidence trail from a single log group per function.
As a premier-tier AWS partner with experience applying generative AI to real analytical workflows, Avahi could move from concept to a working, demonstrable prototype within a defined engagement window. Untamed Planet needed a partner who could design a scoring methodology from scratch, validate it against a labeled reference sample, and hand over a foundation the team could build on for a production version, rather than a one-off proof of a single idea.
Avahi designed and built an end-to-end prototype that takes a submitted campaign, a URL or keyword paired with an organization, and ingests content about it from a confirmed set of public sources through several parallel ingestion adapters. The pipeline captures publicly visible reach and engagement signals where a source exposes them, and falls back to content-based signals, such as volume of coverage and number of distinct sources, where it doesn’t, so every report is clear about which figures are directly measured and which are inferred.
Retrieved reactions are classified into a defined set of sentiment and tone categories, refined over the course of the engagement to sharpen how skepticism is distinguished from ordinary criticism, and the campaign is benchmarked against a comparable organization on the same signals. Each report generates an effectiveness and credibility score using an LLM-as-judge approach across multiple foundation models on Amazon Bedrock, with each model’s assessment paired with a written rationale rather than a bare number, so the reasoning behind each score is visible and reviewable.
A single-page interface lets a user submit a campaign, trigger the analysis pipeline, and review the generated report end to end, with every figure in the report traceable back to the source item behind it. Avahi validated the sentiment approach against a client-provided labeled reference sample spanning several hundred human-labeled reactions, refined the classification categories based on review feedback, and handed over the full source code, configuration, deployment automation, and architecture documentation.
Untamed Planet came out of this engagement with a working, demonstrable prototype that grounds sustainability messaging assessments in retrieved public evidence rather than individual expert judgment alone. The team validated the approach end to end against sample campaigns, reviewed the generated reports, and now has a documented foundation, source code and deployment automation included, to pursue a production build and validate the concept with prospective clients.
Let’s explore your high-impact AI opportunities together in a complimentary session
