Project Overview
Bravo Foxtrot is an innovative security organization dedicated to combating human trafficking and exploitation by identifying perpetrators and vehicles through AI-driven image similarity. They needed a faster, more accurate way to process and compare images from various data sources. Avahi designed and delivered a streamlined proof of concept using AWS to rapidly ingest and analyze images. As a result, Bravo Foxtrot gained a scalable, secure solution that improves global investigative operations and speeds time to insight.
About the Customer
Bravo Foxtrot operates within the security industry, focusing on solutions that expose and halt human trafficking activities. By harnessing emerging technology and AI, they strive to make investigative processes more efficient and effective to help save lives around the world.
The Problem
Bravo Foxtrot’s teams faced massive volumes of unstructured images requiring rapid comparison and classification. Manual methods proved inefficient and risky, as crucial leads could go undetected if not processed quickly. With investigations spanning various regional databases, Bravo Foxtrot needed an automated solution to ensure that critical evidence was never overlooked. Failure to address this challenge would not only slow down rescue operations but could result in missed opportunities to apprehend dangerous perpetrators.
Why AWS
Bravo Foxtrot chose AWS for its unmatched global presence, allowing secure and rapid deployment wherever investigative teams are located. Additionally, AWS’s portfolio of AI and machine learning services —especially Amazon Bedrock and Amazon SageMaker—provided the robust, scalable infrastructure needed for this demanding image-similarity proof of concept.
Why Bravo Foxtrot Chose Avahi
Avahi’s proven track record in designing advanced AI and ML solutions on AWS made them an ideal partner. With deep expertise in AWS services like S3, Lambda, API Gateway, and SageMaker, Avahi was well-positioned to quickly develop a Gen AI-based proof of concept. Bravo Foxtrot appreciated Avahi’s ability to strictly manage timelines and budgets, ensuring the project fit into a three-week engineering window while leveraging AWS POC funding
Solution
Avahi began by establishing a secure data flow, using Amazon S3 to store Bravo Foxtrot’s images for analysis. SageMaker pipelines were created for both feature extraction and inference, enabling quick training and testing of multiple image-similarity algorithms—up to five different models for comparative confidence levels.
AWS Lambda functions powered the preprocessing logic, ensuring data was cleaned and normalized before being passed to the inference pipeline. The solution leveraged API Gateway to allow internal teams to request real-time model results securely. Amazon Bedrock served as the foundation for harnessing generative AI capabilities, while a Pinecone Vector Database stored embeddings to facilitate rapid similarity searches at scale.
Avahi adhered to a structured three-phase delivery approach:
- Discovery & Planning: Defined success criteria, validated data flows, and finalized AWS connectivity.
- Design & Development: Built and tested the image-similarity pipelines, delivered SageMaker notebooks, and created inference jobs.
- Quality Check & Handover: Conducted user testing, ensured accuracy standards aligned with Bedrock default models, and transferred the finalized proof of concept to Bravo Foxtrot.
AWS Services Utilized
- Architecture design documentation
- SageMaker notebooks for feature extraction
- SageMaker inference notebooks demonstrating the image-similarity POC
- Data ingestion process via S3 buckets
- API Gateway endpoints for inference requests
- Preliminary performance testing and validations
Project Impact
By implementing this AWS-based proof of concept, Bravo Foxtrot can now rapidly identify matching images, improving investigative outcomes and accelerating response times. The flexible architecture means Bravo Foxtrot can further expand capabilities to meet evolving needs in their global mission.
- Completed solution within the 3-week POC window
- Integrated up to 5 image-similarity algorithms for comparative insights
- Adhered to the default accuracy levels established by Amazon Bedrock
- Established a fully scalable foundation for future AI innovation
