Skip to content

Latest commit

 

History

History
31 lines (23 loc) · 1.85 KB

README.md

File metadata and controls

31 lines (23 loc) · 1.85 KB

What’s the experiment?

  1. Take N companies with different ATS
  2. Calculate the lowest, average, median, and highest amount of text we need to aggregate those vacancies to become searchable (basically, to generate tags and skills based on that)

Metrics to count

  1. Min., Average, Median and Max. amount of text as input
  2. Min., Average, Median and Max. amount of tokens as input
  3. Daily, weekly, monthly refresh cost with an average update for ~3 times / month

Results

Filter Num. Companies Num. Jobs Min. Max. Avg. Median
Characters 5 361 1391 6956 3463 3567
Tokens 5 361 283 1339 681 659

Converting to OpenAI GPT-3.5 input tokens pricing (0.0010$ / 1000 tokens):

Category Tokens Price Price per 100 vacancies Refresh cost (3x / month) Refresh cost (100 vacancies / month)
Min. 283 0.000283$ 0.0283$ 0.000849$ 0.0849$
Max. 1339 0.001339$ 0.1339$ 0.004017$ 0.4017$
Avg. 681 0.000681$ 0.0681$ 0.002043$ 0.2043$
Median 659 0.000659$ 0.0659$ 0.001977$ 0.1977$

Meaning that, with a budget of 100$ / month, we can approximately hold:

  • Median estimate: ~50600 vacancies in the database in up-to-date state.
  • Highest estimate: ~25000 vacancies in the database in up-to-date state.