Clay Icon

Refining Job Postings: Filtering by Included and Excluded Keywords

·
·

Hey folks, I have a use case around PredictLeads job postings where I'm getting stuck. I want to return Job Postings that match my keyword inclusion list and don't match my exclusion list. why: I use Linkedin Alerts for job opening signals today, but Linkedin is really bad with a) context/num characters it allows b) "NOT" operators. I want to filter out the noise in open job postings, so I can reduce the manual effort here. Here's pseudocode. Included Keywords: Marketing, Data, Digital Excluded Keywords: Crime, Fraud, Security, Sales, Talent Based on Company Domain and Included Keywords, look up open roles via PredictLeads. A JSON object returns, which could contain 15 job openings matches for a given company. 1 - If any of the Excluded Keywords are found in Job Title[0], ignore the role entirely and move onto Job Title[1]. If there is no match, move onto step 2. 2 - See if the Job Title[0] matches any of the Included Keywords. If not, see if the Job Description[0] matches any of the Included Keywords. ----- 2A If not, ignore the role entirely and move onto Job Title[1] ----- 2B If a match is found, write down the sentences where that keyword was found and continue parsing the description for more tidbits. Add all of this into an object/array. 3. Do this again for all x results in that company's open job postings 4. Once I have an object with a slimmed down Job Title > Relevant Sentences, I can begin to build out some subject lines/first lines that might be relevant Problems I'm experiencing:

  • I can't use the AI integration, since the job posting description itself is far too many characters for Clay's integration to handle (on free plan). I tried using the AI Formula Generator and got closer, but still not there. (maybe I parse this description into 3-4 separate fields and run the AI on all of them independently?)

  • I'm having trouble with Clay being able to handle nested JSON objects - it feels easy to de-construct an object but I'm not sure if I can construct an object.

  • Avatar of Clay Team
    Clay Team
    APP
    ·
    ·

    Hi Matt, sorry for the delay here. Would love to take a look at your table -- can you send me a link to it?

  • Avatar of Matt M.
    Matt M.
    ·
    ·
  • Avatar of Clay Team
    Clay Team
    APP
    ·
    ·

    Hey Matt, I'll take a look here too

  • Avatar of Matt M.
    Matt M.
    ·
    ·

    Thanks Arturo O. - I made some updates but it's still not what I want.

  • Avatar of Arturo O.
    Arturo O.
    ·
    ·

    Sorry for the delay, Matt. I forgot to mention what I found since I had a few calls in between... To clarify your need here, did you want to give chatGPT all available descriptions the enrichment returned and only that info so it could best generate it's message?

  • Avatar of Arturo O.
    Arturo O.
    ·
    ·

    I added an Extract Field integration (doesn't use credits), to extract these specific fields, descriptions only, making it easier for chatGPT to only read that list instead of all from the enrichment. You can give chatGPT this specific list if that's what you needed

  • Avatar of Arturo O.
    Arturo O.
    ·
    ·

    Also, the formula generator usually does a very good job with JSON but may need a little bit of prompting so it understands and maps out the correct fields you need as the output. Otherwise, chatGPT has a JSON mode which would achieve something similar -> JSON mode in chatGPT: https://www.loom.com/share/317de47023fb429a80022de0f4a5d6dc

  • Avatar of Matt M.
    Matt M.
    ·
    ·

    Arturo O. What I was hoping for is for ChatGPT to summarize the Job Description, but it doesn't have a large enough context to do so, I think. I like the Extract Field from Object - this is helpful. I'll give it another shot this week

  • Avatar of Clay Team
    Clay Team
    APP
    ·
    ·

    Cool! sounds good! To clarify, the model that you use in regards to chatGPT, each one is limited by openAI itself.So 3.5 turbo can process up to 16k tokens model 4 is down to 8k tokens per prompt model 4-turbo is up to 128k tokens which is huge but rate limits are a bit lower than the other models, in any case, it could probably sum everything up if you select this one as default.If it helps, OpenAI's token calculator: https://platform.openai.com/tokenizer

  • Avatar of Daniel K.
    Daniel K.
    ·
    ·

    Arturo O. haven't had any luck with the formula generator. For you particuar example, can you give me a formula prompt? cheers EDIT: It does the job but the json doesnt show in the output window

  • Avatar of Arturo O.
    Arturo O.
    ·
    ·

    Hey Daniel K., sounds like it worked, right? Yeah, the output window is limited in some regard and may need improvement, but if it's still generating the value after you save, then you can analyze some examples and modify the formula a bit if needed.