Back to Projects
EvoPlay - AI Platform for Evidence-Based Student Interventions

EvoPlay - AI Platform for Evidence-Based Student Interventions

Work
CML Insights•2024 - 2025•Machine Learning Engineering Lead

Technologies

pythonfastapinextjslanggraphopenaipostgresvector-databasesragkubernetesterraformgcpkustomizevault

Key Highlights

  • ▸Principal-level ownership of an AI-driven decision platform for proactive student interventions
  • ▸Designed multi-agent LLM architecture for evidence-based recommendations at scale
  • ▸Led system design across ML, LLM orchestration, backend APIs, and real-time UX
  • ▸Built natural-language interfaces backed by structured reasoning and evidence metadata
  • ▸Advised teams on agent design, RAG, and production-grade AI systems

Overview

AI-powered decision platform that transforms student support from reactive and intuition-driven to proactive and evidence-based. EvoPlay acts as a smart assistant for student success teams—identifying at-risk students early, matching them to proven interventions, and generating personalized, actionable treatment plans.

Architecture

End-to-end, multi-agent AI system:

  • Frontend: Next.js with real-time streaming responses
  • Backend: FastAPI services orchestrating data access and AI workflows
  • AI Layer: LangGraph-based multi-agent system with structured outputs
  • Data: PostgreSQL for student data, vector databases for intervention evidence
  • Delivery: Real-time responses via server-sent events (SSE)

Key Technical Contributions

Intelligent Student Analysis

Designed systems to track student characteristics (engagement, motivation, GPA, preparedness, socioeconomic indicators) and proactively identify at-risk students. Enabled behavioral segmentation to support targeted, high-impact interventions.

Evidence-Based Matching Engine

Built vector-based matching between student profiles and validated interventions. Recommendations incorporate statistical significance, effect sizes, and population alignment—ensuring decisions are grounded in evidence rather than heuristics.

Multi-Agent AI System

Architected a LangGraph-powered agent ecosystem:

  • Router Agent to classify user intent
  • Query Agents to generate and execute SQL automatically
  • Evidence Agents to retrieve and reason over intervention data
  • Recommendation Agents to produce treatment plans
    Responses are streamed in real time and backed by explicit reasoning and metadata.

Actionable Outputs & Explainability

Enabled generation of personalized treatment plans, implementation steps, communication templates, and on-demand plots directly from conversational queries. All outputs include traceable reasoning and supporting evidence surfaced to end users.

Technologies

  • Backend: Python, FastAPI
  • Frontend: Next.js, Tailwind CSS, SSE streaming
  • LLMs & Agents: OpenAI, LangGraph, LangSmith
  • RAG: Vector databases with structured retrieval
  • Data: PostgreSQL
  • Deployment: Kubernetes

Impact

Empowered institutions to identify risk earlier, allocate limited resources effectively, and apply interventions with proven impact. EvoPlay bridges student data, research evidence, and AI-driven reasoning into a single, operational decision platform.