/
SupportXDR
Ontario, Canada
Software (Enterprise AI / SaaS)
Amazon Bedrock, AWS Lambda, Amazon API Gateway, Amazon Aurora PostgreSQL (Serverless), Amazon S3 Vectors, Amazon S3, Amazon CloudFront, Amazon EventBridge, Amazon CloudWatch, Amazon ECR, AWS IAM
SupportXDR is developing Metarri, an AI-native platform that surfaces evidence-backed insights from fragmented enterprise data. In a four-week engagement, Avahi built a working prototype of the platform’s insight layer: three specialized AI agents operating over unified organizational data, feeding a validation dashboard where reviewers see traceable, confidence-scored findings. The system uses Amazon Bedrock foundation models orchestrated by AWS Lambda, with data indexed in Amazon S3 Vectors and stored in Amazon Aurora PostgreSQL Serverless. Delivery closed with a customer satisfaction score of 5 out of 5, and the customer moved directly into scoping a follow-on production engagement.
SupportXDR builds AI-driven software that helps enterprise teams cut through fragmented tools, chaotic dashboards, and scattered information sources. Its Metarri platform is designed to give organizations a single, intelligent view of activity across the systems where work actually happens, from ticketing and collaboration tools to knowledge bases and CRM. The company operates in the enterprise software space, focused on turning disconnected organizational signals into structured, actionable intelligence.
Enterprise teams generate high volumes of signal every day across Jira, Slack, Confluence, Google Drive, CRM, and calendar systems. That signal contains early evidence of risks, opportunities, and knowledge gaps, but it sits siloed across tools, none of which surface cross-system patterns. Teams rely on tribal knowledge or manual review to spot the issues that matter, which means most weak signals are noticed only after they escalate.
SupportXDR set out to build Metarri to solve this at scale, but faced a validation question first: can generative AI reliably produce high-quality, evidence-grounded insights from real enterprise data, or does it collapse into low-value output and hallucination? Confidence in that answer was a prerequisite for further investment. Without it, the platform’s core value proposition of precision and traceability over volume could not be committed to, and any subsequent product build would rest on unproven ground.
SupportXDR chose AWS to develop and validate the Metarri insight layer because AWS offered the ready-to-use generative AI and vector search primitives the architecture depended on, without pulling the team into infrastructure work. Amazon Bedrock provides direct access to leading foundation models under a consistent API, letting agent logic iterate on prompts and models without managing model hosting. Amazon S3 Vectors and Amazon Aurora PostgreSQL Serverless handle semantic similarity and structured storage on managed services that scale with usage, and AWS Lambda gives the team an event-driven runtime that fits both the scheduled batch and on-demand invocation patterns the workload requires.
AWS also gives Metarri a foundation to extend as the platform matures, from swapping between additional Bedrock foundation models and inference profiles, to layering in more source connectors, expanded vector search, and multi-region deployment. The platform can grow within a single ecosystem rather than stitching together disparate providers.
Avahi is a Premier Tier AWS Partner with a track record of taking generative AI concepts from scope to working product on AWS in weeks rather than months. SupportXDR chose Avahi to validate the Metarri insight layer because the engagement required both applied GenAI expertise (multi-agent design, prompt engineering, evidence grounding, confidence scoring) and AWS-native delivery (Bedrock, Lambda, S3 Vectors, API Gateway) under one delivery team.
The customer’s priority was precision over volume: a system that would surface a small number of validated, evidence-backed insights, or clearly report that the data was insufficient, rather than generate confident-sounding noise. That discipline of designing for reliability and traceability aligned with how Avahi structures GenAI engagements.
Because this was SupportXDR’s second engagement with Avahi, the team also brought continuity from the earlier collaboration into scoping, kickoff, and delivery cadence.
Avahi delivered a functioning multi-agent insight system, spanning ingestion, agent execution, an insight validation layer, and a review dashboard, deployed on AWS in the customer’s account.
Data ingestion and shared context layer. Enterprise data covering Jira, Slack, Confluence, Google Drive, CRM, and calendar activity is normalized into a unified schema and stored in Amazon Aurora PostgreSQL Serverless. Records are stored in the PostgresDB, pulled and we built an evidence bundle. Insights ar embedded to avoid duplication. Records in the DB since most of them are not natural language and include useful metadata to sort through when needed.
Three specialized agents on Amazon Bedrock. A single AWS Lambda function orchestrates three agents (Silent Failure Detection, Opportunity Archaeologist, and Knowledge Gap Mapper) that run concurrently over the same evidence bundle. Each agent inherits shared logic and provides only its own name, insight type, and system prompt. Every agent call goes through Amazon Bedrock’s Converse API against a Claude Sonnet foundation model, producing structured findings tagged as risk, opportunity, or gap insights. The Lambda can run in batch mode against all active projects (scheduled by Amazon EventBridge) or against a single project on-demand.
Evidence-first judging and confidence scoring. After the agents generate candidate insights, a judge stage runs against Bedrock to verify each insight’s supporting evidence against the database, drop insights whose evidence cannot be traced, and rescore the survivors’ confidence with reasoning. Semantic deduplication against past insights in S3 Vectors ensures new insights are not near-duplicates of ones the reviewer already saw. This design directly enforces the customer’s precision-over-volume requirement.
Backend, API, and validation dashboard. A FastAPI backend, packaged as a container image and deployed as an AWS Lambda function behind Amazon API Gateway, serves the domain APIs for customers, projects, insights, agent runs, and every source record type. A web-based dashboard, hosted on Amazon S3 and served through Amazon CloudFront, lets reviewers filter by insight type (risk, opportunity, gap), sort by confidence, drill down into cited evidence, run insight generation per project and per agent, and mark insights as open, in progress, or completed. The dashboard authenticates through a JWT layer, and all container images are stored in Amazon ECR with a lifecycle policy retaining the most recent builds.
Post-demo refinement. Following the mid-engagement demo, the team incorporated client feedback into the delivered build, including support for insights tagged with more than one type, an insight status column with active and inactive filtering, structured recommendations (owner, action, reason, timeframe), and improvements to the confidence rationale display and evidence
The engagement produced a working end-to-end prototype of the Metarri insight layer within a four-week window, closed on the agreed timeline, and gave SupportXDR the evidence needed to move toward a production build on the same AWS foundation. The customer completed formal sign-off and immediately opened discussions on productionizing the platform with Avahi.
SupportXDR
Ontario, Canada
Software (Enterprise AI / SaaS)
Amazon Bedrock, AWS Lambda, Amazon API Gateway, Amazon Aurora PostgreSQL (Serverless), Amazon S3 Vectors, Amazon S3, Amazon CloudFront, Amazon EventBridge, Amazon CloudWatch, Amazon ECR, AWS IAM
SupportXDR is developing Metarri, an AI-native platform that surfaces evidence-backed insights from fragmented enterprise data. In a four-week engagement, Avahi built a working prototype of the platform’s insight layer: three specialized AI agents operating over unified organizational data, feeding a validation dashboard where reviewers see traceable, confidence-scored findings. The system uses Amazon Bedrock foundation models orchestrated by AWS Lambda, with data indexed in Amazon S3 Vectors and stored in Amazon Aurora PostgreSQL Serverless. Delivery closed with a customer satisfaction score of 5 out of 5, and the customer moved directly into scoping a follow-on production engagement.
SupportXDR builds AI-driven software that helps enterprise teams cut through fragmented tools, chaotic dashboards, and scattered information sources. Its Metarri platform is designed to give organizations a single, intelligent view of activity across the systems where work actually happens, from ticketing and collaboration tools to knowledge bases and CRM. The company operates in the enterprise software space, focused on turning disconnected organizational signals into structured, actionable intelligence.
Enterprise teams generate high volumes of signal every day across Jira, Slack, Confluence, Google Drive, CRM, and calendar systems. That signal contains early evidence of risks, opportunities, and knowledge gaps, but it sits siloed across tools, none of which surface cross-system patterns. Teams rely on tribal knowledge or manual review to spot the issues that matter, which means most weak signals are noticed only after they escalate.
SupportXDR set out to build Metarri to solve this at scale, but faced a validation question first: can generative AI reliably produce high-quality, evidence-grounded insights from real enterprise data, or does it collapse into low-value output and hallucination? Confidence in that answer was a prerequisite for further investment. Without it, the platform’s core value proposition of precision and traceability over volume could not be committed to, and any subsequent product build would rest on unproven ground.
SupportXDR chose AWS to develop and validate the Metarri insight layer because AWS offered the ready-to-use generative AI and vector search primitives the architecture depended on, without pulling the team into infrastructure work. Amazon Bedrock provides direct access to leading foundation models under a consistent API, letting agent logic iterate on prompts and models without managing model hosting. Amazon S3 Vectors and Amazon Aurora PostgreSQL Serverless handle semantic similarity and structured storage on managed services that scale with usage, and AWS Lambda gives the team an event-driven runtime that fits both the scheduled batch and on-demand invocation patterns the workload requires.
AWS also gives Metarri a foundation to extend as the platform matures, from swapping between additional Bedrock foundation models and inference profiles, to layering in more source connectors, expanded vector search, and multi-region deployment. The platform can grow within a single ecosystem rather than stitching together disparate providers.
Avahi is a Premier Tier AWS Partner with a track record of taking generative AI concepts from scope to working product on AWS in weeks rather than months. SupportXDR chose Avahi to validate the Metarri insight layer because the engagement required both applied GenAI expertise (multi-agent design, prompt engineering, evidence grounding, confidence scoring) and AWS-native delivery (Bedrock, Lambda, S3 Vectors, API Gateway) under one delivery team.
The customer’s priority was precision over volume: a system that would surface a small number of validated, evidence-backed insights, or clearly report that the data was insufficient, rather than generate confident-sounding noise. That discipline of designing for reliability and traceability aligned with how Avahi structures GenAI engagements.
Because this was SupportXDR’s second engagement with Avahi, the team also brought continuity from the earlier collaboration into scoping, kickoff, and delivery cadence.
Avahi delivered a functioning multi-agent insight system, spanning ingestion, agent execution, an insight validation layer, and a review dashboard, deployed on AWS in the customer’s account.
Data ingestion and shared context layer. Enterprise data covering Jira, Slack, Confluence, Google Drive, CRM, and calendar activity is normalized into a unified schema and stored in Amazon Aurora PostgreSQL Serverless. Records are stored in the PostgresDB, pulled and we built an evidence bundle. Insights ar embedded to avoid duplication. Records in the DB since most of them are not natural language and include useful metadata to sort through when needed.
Three specialized agents on Amazon Bedrock. A single AWS Lambda function orchestrates three agents (Silent Failure Detection, Opportunity Archaeologist, and Knowledge Gap Mapper) that run concurrently over the same evidence bundle. Each agent inherits shared logic and provides only its own name, insight type, and system prompt. Every agent call goes through Amazon Bedrock’s Converse API against a Claude Sonnet foundation model, producing structured findings tagged as risk, opportunity, or gap insights. The Lambda can run in batch mode against all active projects (scheduled by Amazon EventBridge) or against a single project on-demand.
Evidence-first judging and confidence scoring. After the agents generate candidate insights, a judge stage runs against Bedrock to verify each insight’s supporting evidence against the database, drop insights whose evidence cannot be traced, and rescore the survivors’ confidence with reasoning. Semantic deduplication against past insights in S3 Vectors ensures new insights are not near-duplicates of ones the reviewer already saw. This design directly enforces the customer’s precision-over-volume requirement.
Backend, API, and validation dashboard. A FastAPI backend, packaged as a container image and deployed as an AWS Lambda function behind Amazon API Gateway, serves the domain APIs for customers, projects, insights, agent runs, and every source record type. A web-based dashboard, hosted on Amazon S3 and served through Amazon CloudFront, lets reviewers filter by insight type (risk, opportunity, gap), sort by confidence, drill down into cited evidence, run insight generation per project and per agent, and mark insights as open, in progress, or completed. The dashboard authenticates through a JWT layer, and all container images are stored in Amazon ECR with a lifecycle policy retaining the most recent builds.
Post-demo refinement. Following the mid-engagement demo, the team incorporated client feedback into the delivered build, including support for insights tagged with more than one type, an insight status column with active and inactive filtering, structured recommendations (owner, action, reason, timeframe), and improvements to the confidence rationale display and evidence
The engagement produced a working end-to-end prototype of the Metarri insight layer within a four-week window, closed on the agreed timeline, and gave SupportXDR the evidence needed to move toward a production build on the same AWS foundation. The customer completed formal sign-off and immediately opened discussions on productionizing the platform with Avahi.
Let’s explore your high-impact AI opportunities together in a complimentary session
