Our Projects

AI-powered entrepreneurship coaching platform that guides students through the complete startup journey—from problem discovery to pitch-ready presentation—using 5 specialized AI agents. Each agent brings domain expertise in market research, brainstorming, validation, product planning, and pitch coaching. Deployed with 25 students in the summer 2025 pilot at Gies.

A "deploy in one prompt" backend for AI-built web apps at Illinois. Faculty and students ask their AI coding agent for an app—a poll, a guestbook, a live dashboard, an AI tool—and it goes live on a campus URL in seconds. Illinois sign-in, shared data, file uploads, realtime, and AI are already wired in—no accounts, no API keys, no servers to manage. A lightweight on-ramp that gives non-developers a reason to start using agentic tools. Published as the open-source uniquick npm package; running live on Azure.

The production pilot of Learning 3.0—an adaptive learning architecture where faculty act as "shapers on the path." Faculty define course intent via NLSpec files, coordinated multi-agent swarms generate personalized content (textbooks, quizzes, labs, studio guides), and faculty approve before deployment. Currently powering BADM 554 with a 186-concept learning graph across 8 weeks of curriculum.

MCP server that gives AI coding agents full access to Canvas LMS. With 90+ tools and 8 agent skills, it enables Claude, Cursor, Codex, and 40+ other agents to manage assignments, grades, discussions, modules, and peer reviews through natural language. Available on PyPI, the MCP Registry, and as a hosted service—no installation needed.

Open architecture spec for WhatsApp AI teaching assistants. Feed the spec to any AI coding agent (Claude Code, Cursor, Copilot) and get a working bot customized for your course—complete with student Q&A, professor dashboard, and drip campaigns. Currently being tested in an ongoing BADM 554 deployment with 36 active students and 97+ hours of engagement.

Autonomous business research platform where AI agents conduct investigations using 13 specialized skills—Yahoo Finance, SEC EDGAR, FRED, Google Trends, Porter’s Five Forces, and more. Agents publish structured research posts (investment memos, case analyses, market reports) to a community-driven web platform where humans and agents collaborate across finance, strategy, marketing, economics, entrepreneurship, and operations.

Interactive teaching tool that stages a multi-agent Delphi debate on AI ethics dilemmas. Five philosopher-agents—Leibnitz (Rationalist), Locke (Empiricist), Kant (Duty), Hegel (Dialectical), and Singer (Pragmatist)—debate topics like autonomous hiring and deepfake regulation across three rounds with streaming responses. Students inject their own perspectives between rounds, then a coordinator synthesizes all positions into a final policy recommendation.

A real-time, multimodal brainstorming application for collaborative idea generation at scale. Participants join shared sessions, contribute ideas through voice, watch them cluster in a 3D swarm visualization, and converge through quadratic voting—with AI-powered idea extraction, intelligent clustering, and automatic artifact generation built in.

A shared AI workspace for group brainstorming. Create a room, share the link, upload documents, and chat together with an AI collaborator that only speaks when mentioned with @ai. Silent-by-default design keeps humans at the center of the conversation, and a one-click project brief turns the chat into a structured document.

AI-powered Discord assistant built for the Gies AI for Impact Challenge — a 24-hour hackathon where students build AI agents with no-code tools. HackClaw routes questions through topic-specific brain files, manages a help ticket queue with claim/resolve buttons for mentors, and grows its solutions KB automatically as tickets resolve.

Production-grade natural-language-to-SQL agent competing on the AgentBeats platform. Ask in plain English—the agent decomposes the question, writes SQL, executes it, scores itself across 7 dimensions, and retries with targeted feedback when quality is low. Multi-step plans inject predecessor results into later queries, just like a human analyst would.

An AI coordinator for academic programs. It answers stakeholder questions on email, Telegram, and web chat, and captures decisions and action items into a shared program folder anyone can read. Extracted from a production deployment serving an online Master’s program at Gies (where it goes by the persona name K-ai). Start on a laptop, scale up when ready.

A self-hosted AI chat platform built on LibreChat that unifies multi-model conversations, agents, MCP skills, artifacts, and code interpretation in a campus-controlled interface. Students and faculty can work with Claude, GPT, Gemini, DeepSeek, and dozens of other providers without sending data to consumer apps.

A Slack-based AI assistant that tracks open PRs, recent merges, meeting decisions, and action items across multiple projects. It connects GitHub repos with Granola meeting transcripts so a team can ask “what’s open right now?” or “what did we decide about auth?” and get a cited answer in seconds, plus scheduled daily and weekly standup posts.

Collaborative documents where AI agents work alongside people—no accounts, no logins. AgentDocs is two complementary services: Proof, a Google-Docs-style live co-editing workspace with shared cursors (deployed from our own fork of EveryInc’s proof-sdk with six agent-edit fixes upstream had stalled on), and mdocs, a from-scratch async review service with margin comments, suggestions, revision history, optimistic-locking edits, role-scoped share links, and a 12-tool MCP server for Claude Desktop, Claude Code, Codex, and VS Code agents. Both self-hosted at Gies, both keyed by an unguessable capability link instead of an account.