Skip to content

Latest commit

 

History

History
63 lines (40 loc) · 2.55 KB

README.md

File metadata and controls

63 lines (40 loc) · 2.55 KB
Raw

ai-agent-recruiter

Building an AI Agent which can talk to candidate on whatsapp, process his resume for a job and conduct a basic interview!

This agent is live, try it out by sending "Hi" at 91-9870413312

If you want to test diffrent flows, send whatsapp message "CLEAR" this will delete all memory of user and start afresh!

Stack

  • Using LLM's (DeepSeekV2, Claude (in-progress))
  • Langfuse for observability
  • Plivo for whatsapp
  • MongoDB for database
  • AssemlyAI/Deepgram for Voice to Text

It will reach out the candidate on whatsapp based on this resume and phone no. Do END-END shortlisting and conduct a basic whatsapp audio based interview

Whatsapp Agent

  • Check if candidate is available by sending a curated message on whatsapp
  • Based on his reply
  • Ask few basic screening round questions like salary, location, etc
  • Based on predefined rules, if all OK
  • Ask basic technical questions to the candidate on whatsapps and ask candidate to reply via message
  • This allows candidate to quickly respond to question and we are able to analyze his communication skills as
  • Shortlist candidate for technical round

Keep sending status to recrutiers on slack

  • Keep candidate communications updated on slack so recruiters can manually view it
  • Recruiters can anytime take over the conversion via slack and stop the automated bot replies

Agent Architecture

Agent has a STATE (s) , STATE TRANSITION (s) and RULE (s)

  • Based on the conversation I extract certain key information regarding by agent using LLMs. this can be seen in extractInfo()

  • Based on this information, we do STATE Transition and move user to a specific state. these states can be seen in rule_map

  • For every state, based on the conversation there are rules defined which allow agent to generated specific output

  • Agent generates a specific output which is sent to the user and also generations an action

  • Based on the action I update state.

  • User responds

  • This LOOP is repeated

Things to work on

  1. Able to ask questions real time and judge candidate on tech skills properly and provide a rating
  2. If candidate are sending message, after status is completed. There should be option for hr to reply manually. they need to get notified about it.
  3. if job opening gets closed, inform previous candidates?
  4. when we copy converstion of one channel to another, sometimes if the conversaion is old media doesn't get downloaded and the process fails.
  • create a cron job to delete 1month old candidate and move to archieve so they can be processed again if they apply