Agentic AI Hackathon, Challenge Repo

This repo contains the official challenge materials for a hands on hackathon focused on building agentic workflows with structured, realistic datasets. Teams will pick a track, follow Stages 1–3 in the track doc, then design an agent based solution that produces a clear triage output, a recommended path to action, and a short adoption plan.

What you build

You will build a working concept that can:

• Read an intake record plus supporting documents
• Normalize messy inputs into a consistent schema
• Identify gaps, risks, and required follow ups
• Route the case to the right step, with a rationale
• Produce a clean summary that a real operator could act on

No coding is required for the hackathon, but teams that want to implement can do so.

Tracks

Each track includes multiple cases and a diverse mix of documents per case. All documents are provided in non binary formats such as txt, csv, and json.

Case packets vary by design, do not assume every case has the same files or completeness.

AMC Track, Academic Medical Center intake triage

Designed for regulated clinical and research adjacent workflows. Typical decision themes include:

• PHI involvement and minimum necessary handling
• IRB readiness and consent considerations
• External vendors and collaborators
• Data security, audit, and access controls
• Routing to compliance, privacy, security, or research operations

EDU Track, Student success intervention orchestration

Designed for campus student support and intervention coordination. Typical decision themes include:

• Academic risk signals and confounders
• Cross office routing, advising, tutoring, financial aid, wellbeing
• Data use constraints and role based sharing
• Outreach channel rules and opt in approaches
• Escalation triggers that require human review

Research Track, Research compliance and intake triage

Designed for research enablement with governance guardrails. Typical decision themes include:

• Human subjects determination and IRB pathway
• Limited dataset handling and DUA requirements
• Cross border collaboration and export control screening
• Biospecimen governance and consent scope
• Tooling requests and vendor security posture

Challenge docs

Start here. The challenge docs and official track instructions are in challenge-docs.

• Shared overview
  Shared rules, constraints, judging focus, and API tool definitions.

• AMC track
  AMC intake triage scenarios, required outputs, and evaluation criteria.

• EDU track
  Student success routing scenarios, rules, and escalation expectations.

• Research track
  Research intake and compliance triage scenarios, governance, and vendor review expectations.

Optional: API connectivity smoke test

• test-api-prompt.md
  A minimal system prompt used to verify the Blob Storage API connections work end to end.

Use this when:

• You just created or changed the API connections (blob_list_blobs_container, blob_get_blob_contents)
• You are troubleshooting access, permissions, or unexpected empty results
• You want to confirm the agent can list blobs and retrieve a known file (example uses criceagenthackresearch and assets/research_case_index.csv)

Suggested workflow for teams

1. Pick a track and read the matching track doc in challenge-docs/.
2. Complete Stages 1–3 in that track doc (tool access → full case packet → case summary contract).
3. Use the track’s case index CSV in assets/ to enumerate case IDs and locate the correct case container and files.
4. Expand into the full bundle requirements (A–F) and run across all cases.
5. Prepare a short demo plus an adoption plan aligned to outcomes.

Output expectations

A strong submission is:

• Consistent across cases
• Explicit about assumptions and missing data
• Clear about next actions and who owns them
• Conservative with sensitive data, aligned to governance
• Easy for a human operator to trust and execute