Requirements are numbered. Functional begin with A, non-functional (qualitative) with Q. Some are based on Assumptions.
This is an iterative process: When deleting, do not reorder or reassign numbers but just leave a gap. Also make sure to maintain the uniqueness of the numbering when adding new entries.
Also consider the research:
- R1: Candidate can create an account (register)
- R2: Candidate can upload a resume (register)
- R3: Candidate can receive tips on their uploaded resume
- R4: Candidate can submit their resume, entering it to the matching process
- R5: Candidate can view matches between their resume and open roles
- R6: Candidate can pull back their resume, removing it from the matching process
- R7: Employer can create an account (register)
- R8: Employer can enter and edit company information such as communication channels, legal and billing
- R9: Employer receives automated help (f.ex. autofill) when entering company information
- R10: Employer can enter, edit and remove open job descriptions
- R12: A story, which is an anonymized version of a resume, is created from a submitted resume
- R13: Employer can read the stories of resumes that match with their open roles.
- R14: Candidates can view matches and read corresponding open roles and employers.
- R15: Employer can unlock the candidate of a match to receive their resume \ full profile (Data Point 2)
- R16: Employer is billed for unlocking a candidate.
- R17: Employer may mark a candidate as hired, or pass on a candidate (Data Point 1)
- R18: Employer is asked to complete a 5-question survey about an unlocked employee (Data point 4)
- R19: Candidate is asked to complete a 5-question survey about the open position and employer of an unlocked match (Data Point 3)
- R20: Candidate can view open roles and employer information of their resume's matches
- R21: An unlocked resume can be transferred to the registered HR system
- R22: Employer can configure HR system for their account (see also R26 & R27)
- R28: Employer receives a monthly report with analyses based on the defined 'Data Points' and further KPIs or metrics
- R30: When a match is marked hired or passed on, demographic data (sex, race, age, and more..) is accumulated (Data Point 5)
- R31: Hiring manager is presented with a dashboard and workspace
- R32: Employee can enter their demographic information (sex, race, age) on the platform
- R11, based on A1: Employer can open accounts for hiring managers.
- R23: based on A1: A hiring manager has access to a subset of an employers open roles and their matches
- R24: based on A3: Candidate can delete their resume from the platform
- R25: based on A12: The candidate must be able to confirm or deny their hired status
- R26, based on A4: An employer can configure several HR systems for transfer
- R27, based on A5: An employer can receive a resume without the configuration of an HR system (f. ex. Email or Download-Link)
- R29: Admin can manually add findings to reports such as statistics and evaluations, that were generated by side-channels (f.ex. DEI consultants shadowing employer interviews).
- R30: Communication with external LLM needs to support 60s delays.
- Q1: The resume tips (R3) must be AI generated
- Q3: The automation when entering company information (R9) must be AI generated
- Q3: Stories must contain a human-readable representation of resume information (S.M.A.R.T (Specific, Measurable, Achievable, Relevant, and Time-Bound) goals, qualifications, and experience), as well as a condensed representation for quantified comparison with open roles.
- Q6: The creation of a story (R12) must use AI
- Q7: The monthly report for an employer conveys possible disparities between those who are hired and those who were not selected.
- Q8: A match between resume and open role must imply similarity / fit
- Q9: Story must be free from racial, lifestyle, and cultural (and potentially further) indicators that lead to unfair biases.
- Q11: The AI is a trained LLM
- Q13: Ease of use: The customer facing UI needs to be attractive and responsive.
- Q2, based on A2: Concurrent activity of several hiring managers does not lead to corrupt or faulty data.
- Q5, based on A8 and A9: AI, when matching or evaluating a resume, must only consider its story (anonymized representation)
- Q12, based on A25: To avoid cost LLMs should not be used only where needed and with intention.
- Q14, based on A10, A27, A28, A20: Change of prompts or AI models need to work without significant decay of matching performance.
- Q15, based on A9, A10, A28: Change of prompts or AI models should reprocess the scoring for matches.
- Q16, based on A31: The number of expected configurations for HR systems are in the thousands, up to 200'000.
- A1: An employer may have several hiring managers representing different branches or departments in the company.
- A2: Hiring managers often work together on the same set of open roles.
- A3: Clear view wants to convey a particularly high level of privacy and security
- A4: Some big employers maintain different HR systems for different branches, departments.
- A5: Some small employers have no common HR system or none at all
- A6: Hiring managers prefer a text about a person instead of stories must be human-readable and convey information about a candidate
- A7: Decisions made by AI are especially difficult to comprehend and justify.
- A8, based on A7: Existing LLMs contain all kinds of biases, including those ClearView seeks to eliminate.
- A9: A fair hiring process seeks to increase determinism and reduce arbitrariness.
- A10: AI models can be deterministic, but training leverages randomness. Different versions of models therefore quickly lead to different results.
- A11: After an unlock, the hiring process will typically continue and conclude outside from ClearView.
- A12: There are malicious or ambivalent employers
- A13: There are malicious or ambivalent candidates
- A17: The upper bound of role openings per month in the US is 250'000, based on these statistics https://www.statista.com/statistics/200003/number-of-job-openings-in-the-us-information-sector/#:~:text=Number%20of%20monthly%20job%20openings%20in%20the%20U.S.%20information%20sector%202005%2D2023&text=As%20of%20May%202023%2C%20the,than%20in%20the%20previous%20years Assuming reaching a market share of 10%, we need to cater to 25'000 openings per month and around the same number of active candidates.
- A18: The user interaction patterns with the platform are sparse (sporadic use) and with light data loads (mainly text).
- A19: The candidate has a need to get information on their statistic
- A20: Due to the quickly evolving AI market, existing LLMs are outmatched soon by future models.
- A21: Specialized and therefore expensive know-how is required to design the use of LLMs in a platform such as ClearView.
- A22: The transmission delays of answers from external LLMs can be very long. 60s are possible.
- A23: External LLM services impose hard rate-limits https://platform.openai.com/docs/guides/rate-limits
- A24: The cost of LLM prompts (operational and during development and testing) varies greatly with prompt size and model choice.
- A25: When including testing, AI know-how, and operational cost, LLMs are the biggest technical cost driving factor for ClearView.
- A26: Funding of Non-Profit projects is at times difficult especially if the estimation of cost varies highly.
- A27: LLM models of external systems will reach end-of-life.
- A28: Different prompts usually result in different output (even if temperature setting is set very low).
- A29: There is a variety of around 100 different HR systems.
- A30: There are 100'000 IT employers as potential clients of the platform.
- A31: With A30, if we also assume that on average, the companies have 2 HR systems, there could be a need of up to 200'000 HR systems being configured.