Polyrific AI Hackathon, Gen 0

Challenge Areas

1. Chat with Data or Documents - Build the most insightful LLM-powered solution for a business use case that can accept one or many documents, one or many datasets, and potentially other inputs, and then initiate an accurate chat session involving the data. Judging criteria include accuracy, utility, cost to operate, and data privacy.
2. PII Redaction - Develop a solution using LLM or other techniques to automatically identify, redact, and then restore PII. Success is measured by redaction/restoration accuracy. PII info will have to be restored to the text generated by LLM with the highest accuracy, ideally 100%.
   1. Base judging criteria:
      1. The solution should be able to identify and flag PII including full names, birth dates, social security numbers, credit card numbers, etc.
      2. The solution should remove the PII from the input text that is to be sent through a language model.
      3. The solution should then receive text from a language model after processing and re-insert the text into the correct place so that the PII is once again visible.
3. LLM-Accelerated Coding - Use LLM to generate code from requirements and docs. Looking for high agency -- can read requirements and create code / read API docs from URL and create code. In this sense, coding begins with a tech lead producing prompts and pointing the agent to a Jira or API doc URL. Assess code quality, functionality, and development time savings.
   1. Base judging criteria:
      1. The solution can create a working code based on the user prompt.
      2. The solution can create quality code with property abstraction, organization, and documentation (in either readme or inline comment).
      3. The solution can retrieve information from external documentation to get the latest/updated solution.
   2. Hackathon Challenge:
      1. The solution can ask for clarification on the prompt and/or the external documentation.
4. Classification - Build classifiers for datasets that are retrieved from any Open Dataset providers. Judged on accuracy, efficiency, and operational costs. Private ANNs or SLMs will be highly valued here.
   1. Base judging criteria:
      1. The solution accepts labeled data for training purposes.
      2. The solution accepts a CSV of emails (format to be provided) and can properly classify them into various categories.
         • Labeled training data will be provided on day 1 of the hackathon so be sure that you have made a method for quickly ingesting training data for your ANN, etc.
         • Image classification is possible in the wildcard round.
5. Anomaly Detection - Detect anomalies in datasets and document sets that are retrieved from any Open Dataset providers. Prioritize performance, cost, and privacy.
   1. Base judging criteria:
      1. Accuracy: Correctly identified anomalies vs. false positives.
      2. Insightfulness: Can the solution provide useful insights from the detected anomalies?
      3. Performance: What’s the processing time? Can it handle scalability when dataset increases?
      4. Cost: What’s the amount of computational resource to get the optimum solution?
6. Evolutionary Algorithm - Optimization challenges like scheduling, cloud spending, logistics, and resource allocation.
   1. Base judging criteria:
      1. The solution can access AWS, GCP, and Azure “retail” APIs that provide SKUs and pricing:
         • AWS
         • Azure
         • GCP
      2. The solution can accept a resource configuration (JSON/XML) and compute the most economical multi-cloud arrangement. For example, if database instances are less expensive on AWS for a given day, app services are less expensive on Azure, and serverless functions are less expensive on GCP, then the evolutionary algorithm should recommend the least expensive resources. The lowest cost is the fitness metric.
      3. The solution should make the recommendation for the lowest cost configuration in a readable report. It does not need to be “pretty”, just readable.

Contestants can start developing solutions that meet the base judging criteria immediately using any resource that they see fit but should be aware we will release surprise requirements the day of the hackathon that they will need to adapt to. This format allows for preliminary work while testing adaptability during the event itself.