Rationale:

• Domain Integration: By focusing on workflows commonly used within and across multiple real estate asset classes the competition ensures that participants must understand the nuances of commercial real estate.

• Technical Acumen: Crafting effective prompts requires a deep understanding of LLM capabilities and limitations, encouraging creative approaches and iterative refinement.

• Industry Innovation: Bridging the gap between domain expertise and AI enables new applications, streamlining processes impacting the work of millions of people around the world.

Competition Structure and Phases

A. Asynchronous Competition

A six week period will be allowed to develop and submit entries

• Task Distribution: Participants will be given a set of commercial real estate workflows, with each set relevant to a particular asset class.

• For instance, an ‘Retail Leasing Workflow’ might include:

◦ Goal: Find the best tenants for a retail space.

◦ Tasks:

1. Identify ideal tenant profiles.

2. Analyse foot traffic data.

3. Compare market rents.

4. Generate outreach emails.

5. Summarise findings for stakeholders.

• Deliverables:

◦ A set of prompts for each sub-task, with annotations explaining the rationale behind each prompt’s design. The workflow might require just one prompt, or a series of iterative prompts. The key is which will be most effective and efficient.

◦ A brief outline of how the prompt(s) address both domain specifics and technical requirements.

• Evaluation: Judges (comprising commercial real estate experts and prompt engineering specialists) review submissions against a detailed rubric (see Section 4).

B. Live Competition

• Live Challenge: In a timed, live session, finalists craft prompts for an unannounced workflow scenario. This tests their on-the-spot thinking and ability to apply their knowledge under pressure.

• Team vs. Individual Options: Entrants can be individuals, or teams of up to four people.

• The ‘Wildcard’ Challenge: Individuals or Teams can come up with their own workflow they’d like to address.

For this challenge we are looking for novelty, imagination and inspiration.

3. Task Design and Example: Leasing Workflow

Using the ‘Retail Leasing Workflow’ example as a template, each task can be designed to test different aspects of the prompt engineer’s skills:

Task 1: Identify Ideal Tenant Profiles

• Domain Knowledge: Requires participants to incorporate factors such as customer demographics, brand alignment, and local market trends.

• Prompt Elements: Asks for a prompt that instructs the LLM to consider specific criteria (e.g. footfall patterns, competitor presence, and complementary retail categories).

Task 2: Analyse Foot Traffic Data

• Technical Focus: Requires participants to design a prompt that not only asks for data analysis but also handles structured data inputs (CSV, JSON) and visualisation outputs.

• Innovation: Encourages the inclusion of statistical analysis, such as peak hours identification and segmentation of traffic by demographics.

Task 3: Compare Market Rents

• Analytical Depth: The prompt would need to guide the LLM to extract and compare data from multiple sources, highlighting regional variations and temporal trends.

• Data Integration: Would test the participant’s ability to blend qualitative insights with quantitative comparisons.

Task 4: Generate Outreach Emails

• Creative Communication: Participants would need to craft a prompt that generates persuasive, tailored outreach emails that align with the identified tenant profiles.

• Language and Tone: Participants would need to evaluate how well the prompt directs the LLM to adopt a professional yet engaging tone, suitable for the commercial real estate industry.

Task 5: Summarise Findings for Stakeholders

• Conciseness and Clarity: The prompt would need to instruct the LLM to compile a concise report summarising key insights, actionable recommendations, and data visualisations.

• Stakeholder Communication: The prompt would need to place emphasis on producing a narrative that resonates with non-technical stakeholders while retaining technical accuracy.

4. Evaluation Criteria and Scoring Rubric

A detailed, multi-dimensional rubric, with weighted scores across several categories:

1. Domain Knowledge (30%):

◦ Accuracy and depth in commercial real estate context.

◦ Relevance of criteria and data considerations.

2. Prompt Engineering Skill (30%):

◦ Clarity, precision, and structure of the prompts.

◦ Technical efficacy in instructing the LLM (e.g. handling of data formats, chain-of-thought strategies).

3. Creativity and Innovation (20%):

◦ Novel approaches, such as chaining prompts or integrating external data sources (in theory via Master

Prompts)

◦ The ability to design prompts that go beyond the obvious to solve problems in unique ways.

4. Performance and Outcome (20%):

◦ Measurable success based on the quality of the LLM’s outputs when using the prompts. (If requiring external data, participants can provide sample, synthetic data sets)

◦ Efficiency, reproducibility, and clarity in final responses.

Additional Considerations:

• Documentation and Rationale: Clear explanations of choices and expected outcomes.

• Adaptability: Ability to handle variations in data inputs or scenario changes.