The Milan AI Week Hackathon brought together developers, founders, and AI innovators to build practical solutions that address real business challenges using modern AI infrastructure. Hosted on Vultr cloud infrastructure, participants focused on moving beyond experimentation to create applications that automate operations, improve decision-making, and help organizations work more efficiently.
Among the standout projects were three winners that demonstrated the power of agentic AI systems, real-time workflow orchestration, and enterprise-ready automation: Cascade AI, ContextGuard, and Deals Machine.
Deals Machine: Keeping sales reps in control with AI-assisted selling
Sales professionals spend significant time researching prospects, updating CRM systems, preparing outreach, and documenting conversations. For smaller sales teams and solo founders, these administrative tasks often reduce the time available for customer engagement.
Deals Machine was created to streamline the sales process while keeping human sellers firmly in control of customer interactions.
The platform serves as a human-driven, AI-assisted cold-calling cockpit that helps sales teams source leads, enrich prospect data, generate personalized outreach strategies, and receive live coaching during conversations. Rather than acting autonomously, the AI provides recommendations and drafts follow-up communications for review, ensuring that sales representatives remain responsible for all customer-facing decisions.
Real-time coaching powered by secure AI workflows
Deals Machine uses a two-tier architecture optimized for low-latency voice processing and secure AI-assisted sales operations.
The user-facing cockpit is built with Next.js and hosted on Vercel, while core processing services run on hardened Fastify workers deployed on Vultr Ubuntu 24.04 infrastructure.
The platform orchestrates lead sourcing, outreach preparation, live coaching, and post-call analysis through a centralized AI layer. Speechmatics provides real-time transcription and speaker diarization, while Anthropic Claude models generate outreach playbooks, deliver coaching recommendations, and classify content for safety.
Twilio connects phone calls directly into the platform, enabling real-time analysis during customer conversations. Supabase stores prospect context, call history, industry guidance, and the platform’s evolving memory system.
One of the project’s most innovative features is its self-improving memory layer, which continuously learns from sales outcomes to refine coaching recommendations and outreach strategies.
The team also prioritized AI security and compliance. Veea Lobster Trap middleware scans third-party content before it reaches language models, helping protect against malicious content and prompt injection attacks originating from scraped data, emails, and business documents.
Cascade AI: Automating customer onboarding from contract to kickoff
Customer onboarding is one of the most critical phases of any client engagement, yet many agencies, consultancies, and software development firms still rely on manual processes to move new customers from signed contracts to active projects.
Cascade AI was built to eliminate these bottlenecks through automated onboarding workflows.
The platform begins when an account manager uploads a signed contract and deal brief. AI agents then analyze contract details, research the customer, generate onboarding plans, assign milestones, and create personalized communications. As customers submit additional information, workflows automatically update without requiring manual intervention.
By automating onboarding operations, Cascade AI helps organizations reduce administrative overhead, accelerate project readiness, and deliver a more consistent customer experience.
Agentic architecture for persistent workflow execution
Cascade AI leverages a persistent AI agent architecture running on Vultr cloud infrastructure to coordinate onboarding activities across multiple business functions.
The workflow begins with a Research Agent powered by Gemini 2.5 Flash, which uses native PDF understanding and URL Context capabilities to analyze contracts and gather information from customer websites. An Orchestrator Agent manages workflow execution and coordinates actions across specialized agents.
Additional agents handle communication and project planning tasks. The Communication Agent generates onboarding emails and customer-facing updates, while the Project Agent builds milestone-based onboarding plans, allocates resources, and identifies potential delivery risks based on contract terms and intake responses.
By combining multimodal AI processing, persistent workflows, and autonomous orchestration, Cascade AI demonstrates how agentic systems can transform operational processes that traditionally require significant manual coordination.
ContextGuard: Real-time validation for operational decision-making
Organizations frequently make important decisions during meetings without complete visibility into prior commitments, legal constraints, project dependencies, or organizational ownership structures. These conflicts often emerge only after work has already begun.
ContextGuard addresses this challenge by acting as a real-time AI meeting agent that validates live discussions against organizational knowledge.
During meetings, the platform continuously analyzes conversations and compares proposed decisions against a knowledge graph containing legal reviews, historical decisions, customer commitments, dependency chains, and ownership information. When conflicts are detected, ContextGuard surfaces evidence-backed alerts and recommended actions before decisions are executed.
The result is improved operational alignment, stronger governance, and fewer costly miscommunications across teams.
AI-powered operational memory and workflow orchestration
Built as a unified Next.js application on Vultr cloud infrastructure, ContextGuard combines real-time speech processing, organizational memory, and AI-driven workflow automation.
Speechmatics provides live transcription with speaker diarization, while Gemini 2.5 Flash extracts commitments, action items, and ownership changes as conversations unfold.
The platform’s organizational memory graph serves as a continuously accessible source of truth, allowing ContextGuard to validate decisions against existing knowledge in real time. When risks or conflicts are identified, users receive intervention cards containing supporting evidence and recommended corrective actions.
Following each meeting, AI agents generate personalized summaries and automate follow-up actions across tools, including GitHub, Jira, Notion, and Gmail. Structured Gemini outputs combined with a Valkey-backed asynchronous action queue allow workflows to continue seamlessly without disrupting the meeting experience.
ContextGuard showcases how AI can move beyond passive note-taking to actively improve organizational decision quality and operational alignment.
Building the future of enterprise AI
The winning projects from the Milan AI Week Hackathon highlight a growing shift in AI development. Rather than focusing solely on chat interfaces or isolated AI features, builders are increasingly creating agentic systems that integrate directly into business operations and continuously execute workflows across organizations.
Whether automating customer onboarding, validating operational decisions in real time, or enhancing sales productivity through human-centered AI assistance, these projects demonstrate how modern AI infrastructure can support practical, enterprise-focused innovation.
By leveraging scalable cloud infrastructure, persistent AI agents, and real-time workflow orchestration, Cascade AI, ContextGuard, and Deals Machine offer a glimpse into the next generation of AI-powered business applications.