Multi-Agent AI SAAS Platform for education

Project Overview

Soria.ai is a multi-agent AI platform designed to automate and orchestrate complex operational workflows across education and support systems. The goal was not to build a single “smart chatbot,” but to create a network of specialized AI agents capable of collaborating, escalating, and integrating with real operational tools.

I led the product vision and delivery of the platform, focusing on reliability, control, and real-world usability across systems like Microsoft Teams, Freshdesk, and Teameo (SIS).

Problem Statement

Organizations were facing:

• Fragmented workflows across chat, ticketing, and student information systems

• Manual triage and repetitive support tasks

• Single-agent bots that failed when conversations became complex

• Lack of control, escalation, and accountability in AI-driven support

There was a clear need for an agentic system that could reason, collaborate, and act within defined boundaries.

Solution

I helped design and deliver a multi-agent AI architecture where each agent had a clear role, responsibility, and decision scope, and could:

• Communicate with other agents

• Decide when to act vs escalate

• Interact directly with operational tools

• Stay aligned with product and business rules

The platform behaves more like a digital operations team than a chatbot.

System Architecture & Approach

The platform was built around agent orchestration and controlled autonomy:

• Specialized agents (triage, support, student info, escalation)

• Defined communication patterns between agents

• Rule-based and context-based escalation flows

• Tool-calling via APIs (Teams, Freshdesk, SIS)

• Guardrails to ensure accuracy, consistency, and safety

Agents were designed to operate independently, but never in isolation.

Key Responsibilities

• Led product vision and roadmap for the multi-agent platform

• Defined agent roles, responsibilities, and collaboration patterns

• Designed automation workflows for ticketing, student management, and support

• Specified escalation logic between agents and to human operators

• Managed backlog, wrote user stories and acceptance criteria

• Validated AI-driven automations against real operational scenarios

• Ensured controlled behavior, reliability, and accuracy of LLM agents in production

Tools & Technologies Used

• LangChain / LangGraph for agent orchestration

• Microsoft Teams API for conversational entry points

• Freshdesk API for ticketing automation

• SIS API for student data and workflows

• Moodle API for course content data

Challenges & Learnings

• Multi-agent systems fail fast without clear role boundaries

• Escalation logic is as important as response quality

• Reliability comes from process design, not model size

• Human-in-the-loop remains essential for edge cases

This project reinforced that agentic AI is an operational discipline, not just an AI pattern.

Outcome & Impact

• Automated large parts of support and student management workflows

• Reduced manual triage and response time

• Improved consistency and traceability of AI actions

• Delivered a scalable foundation for future agents and use cases