CyberDerby: Solo AI hackathon organized by UCSI Information Technology Club (UITC) — the tech club of UCSI University Bangladesh Campus.
The Challenge — NovaPay SOC
You are an AI engineer at NovaPay Bank. Ten security incidents just hit
the queue. Build an autonomous AI agent that reads internal policy documents,
threat intelligence, and employee records — then triages each incident
with a severity rating, containment plan, and analyst briefing.
📦 Download the full challenge pack from the GitHub link in Resources section.
Requirements
✅ Upload code.zip (zip of your code/ folder only — exclude venv/, __pycache__/, data/, dataset/)
✅ Upload output.csv (your agent's triage decisions — 10 rows, exact schema)
✅ Upload log.txt (your full AI session chat transcript from your coding tool)
✅ Short description (max 200 words): explain your agent's approach and AI tools used
Prizes
Champion (1st Place)
Champion Trophy/Crest, Accredited Certificate of Excellence, Prize Money, and fast-track recruitment/internship visibility with MindSpark & Sigmind.
1st Runner-Up (2nd Place)
Runner-Up Crest, Accredited Certificate of Merit, Prize Money, and direct technical recognition from official AI partners.
2nd Runner-Up (3rd Place)
Runner-Up Crest, Accredited Certificate of Merit, Prize Money, and honors at the Grand Award Ceremony on 24 November 2026.
Judges
Abdullah Al Noman
President
Sharhee Saad Imam
Vice President
Abrarul Haque
Judge
Judging Criteria
-
Criterion 1: Accuracy & Schema Compliance (30%)
Exact compliance of output.csv with required schema, column orders, and enums. Correct classification of severity, category, triage decisions across benchmark incidents, and robust detection of adversarial injection cases. -
Criterion 2: Eval & Pipeline Logic (30%)
Clean codebase in code/agent.py, local corpus ingestion with zero hardcoding, rigorous local evaluate.py checks, and complete operational reports (evaluation_report.md custom edge cases & usage_report.md token/cost tracking). -
Criterion 3: AI Innovation & Orchestration (20%)
Effective orchestration of LLMs, structured prompting, multi-step agent reasoning, policy handbook cross-referencing, and model benchmarks. Demonstrated agent engineering rather than naive brute-force prompting. -
Criterion 4: Technical Viva & Session Log Audit (20%)
Live 1-on-1 oral defense (3–5 PM). Complete code mastery, defending design choices line-by-line, zero hardcoded shortcuts, and verification of an authentic AI development trace in log.txt.
Questions? Email the hackathon manager
