Agentic AI running in production on AWS — and, before that, AI for energy, water and finance.
Accenture
nowMulti-agent assistant with RAG over 1,000,000+ documents, running on Bedrock AgentCore.
- CloudFront
- Cognito
- API Gateway
- Orchestrator
- Specialist agents
- Knowledge Base
Designed, built and deployed end to end: an onboarding agent on a serverless AWS architecture. A React frontend served through CloudFront and protected by AWS WAF, with federated SAML 2.0 sign-in through Amazon Cognito.
An API on API Gateway and AWS Lambda feeds a multi-agent system: an orchestrator that calls specialist agents following the Agent as Tool pattern. Behind them, a RAG system runs vector search over a Knowledge Base of more than 1,000,000 documents, indexed from three sources: Jira, web pages and Markdown files.
The agents run containerised on Amazon Bedrock AgentCore, with AgentCore Memory for conversational memory and an evaluation setup that measures how the multi-agent system performs.
Extracts, classifies and cross-checks electrical project documents with AI, at 96% accuracy.
- S3
- Step Functions
- Extract · Classify · Validate
- Cross-check
A serverless solution on AWS that uses AI to extract, classify and validate the documentation of electrical engineering projects.
Pipelines built with AWS Lambda, Step Functions and Amazon S3 apply business rules and cross-document checks to catch inconsistencies between documents.
It reached 96% accuracy extracting and validating 21 documents from low-voltage (BT) and underground low-voltage (LSBT) line projects.
Multi-agent system that analyses, summarises and prioritises case files against ~20 criteria.
- Schedule
- S3
- Step Functions
- Analyse · Summarise · Prioritise
- Web UI
A cloud platform for intelligent case-file processing: a multi-agent AI system analyses, summarises and ranks each file by priority against roughly 20 evaluation criteria, with 5,000 documents analysed.
Asynchronous processing on AWS with S3, Step Functions and scheduled runs, containerised with Docker, and the results surfaced in a web interface.
QUANTIA
Agentic AI for a Tier 1 bank
Support chat for the instant recovery of denied payments, across international operations.
PREDISOLIA
R&D platform for predictive maintenance of photovoltaic plants, built on new AI algorithms.
AQUAHUB
Predictive analytics and advanced monitoring with hybrid AI models for biomethane digesters.
OPTIMA-DONES
Cyber-physical system for proactive monitoring and maintenance of the critical systems of IFMIF-DONES.
The full timeline, dates included, lives on LinkedIn.