workflow.yaml

name: Personalized Research Paper Recommendation Assistant

input_schema:
  type: object
  properties:
    topics:
      type: array
      items:
        type: string
      description: The topics to search for.
    user_id:
      type: string
      description: The user id to search for.

tools:
                          
- name: brave_search
  type: integration
  integration:
    provider: brave
    setup:
      api_key: "DEMO_API_KEY"
      
- name: arxiv_search
  type: integration
  integration:
    provider: arxiv
                          
- name: llama_parse_scrape
  type: integration
  integration:
    provider: llama_parse
    setup:
      llamaparse_api_key: "DEMO_API_KEY"
      params:
        result_format: "text"
        num_workers: 3

- name: get_user_from_id
  description: Get a user from the user id
  type: system
  system:
    resource: user
    operation: list

- name: list_user_docs
  description: List the user docs
  type: system
  system:
    resource: user
    subresource: doc
    operation: list

- name: get_user_docs
  description: Get user docs
  type: system
  system:
    resource: user
    subresource: doc
    operation: list
                          
- name : create_agent_doc
  description: Create an agent doc
  type: system
  system:
    resource: agent
    subresource: doc
    operation: create
                          
- name: create_user_doc
  description: Create a user doc
  type: system
  system:
    resource: user
    subresource: doc
    operation: create

main:

# Calling brave search for each topic
- over: _.topics
  parallelism: 2
  map:
    tool: brave_search
    arguments:
      query: "'the latest news about ' + _"

# Prompting the agent to extract keywords from the news snippet
- prompt:
  - role: system
    content: |
      You are a research assistant.
      Based on the following latest html news snippet, generate keywords:
                        
      {% for docs in _ %}
        {% for doc in docs['result'] %}
          {{doc['snippet']}}
        {% endfor %}
      {% endfor %}
                          
      Your response should be a list of keywords, separated by commas. Do not add any other text.
      Example: `KEYWORDS: keyword1, keyword2, keyword3`
  unwrap: true

# Extracting the top 8 keywords
- evaluate:
    keywords: str(_).split(',')[:8]

# Calling arxiv search for each keyword
- over: _.keywords
  parallelism: 4
  map:
    tool: arxiv_search
    arguments:
      query: _
      max_results: "2"

# Get the arxiv results
- evaluate:
    arxiv_results: _
                          
# Get the user from the id using metadata_filter (returns a list)
- tool: get_user_from_id
  arguments:
    limit: "1"
    sort_by: "'created_at'"
    direction: "'desc'"
    metadata_filter:
      ppid: inputs[0]['user_id']

# Unwrap the list to get the user
- evaluate:
    user_info: _[0]

# Get the user persona from user docs
- tool: get_user_docs
  arguments:
    user_id: inputs[0]['user_id']
    limit: "1"
    sort_by: "'created_at'"
    direction: "'desc'"

# Unwrap the list to get the user persona
- evaluate:
    persona: _[0].content[0]

# Prompting the agent to select the top 5 papers
- prompt:
  - role: system
    content: |
      As an expert in recommending research papers tailored to individual user personas, your task is to select papers top 5 papers that align closely with the user's interests and professional needs. 
      This careful selection ensures that the recommendations are both relevant and beneficial to the user's ongoing research and professional development.

      Below is a detailed description of the user's persona:
      {{_.persona}}

      Based on this persona, please review the following list of research paper titles. 
      These titles have been gathered from a comprehensive search to match the user's interests:

      {% for result in outputs[4].arxiv_results %}
        {% for doc in result['result'] %}
          - (Title: {{doc['title']}}, 
              
          - URL: {{doc['pdf_url']}})
        {% endfor %}
      {% endfor %}

      After reviewing the titles, please compile a list of the top 5 relevant research papers. 
      Your response should be a list of Titles and URLs of the papers, separated by double commas. 
      Do not add any other text.
      Example: `Title1,url1,,Title2,url2,,Title3,url3`
  unwrap: true

# Extracting the titles and urls
- evaluate:
    paper_titles_and_urls: |-
      list(
        zip(
          [pair.split(',')[0] for pair in str(_).split(',,')],
          [pair.split(',')[1].replace("http://", "https://") for pair in str(_).split(',,')]
        )
      )
                          
# Calling llama parse scrape for each url
- over: _['paper_titles_and_urls']
  parallelism: 5
  map:
    tool: llama_parse_scrape
    arguments:
      file: _[1]
      filename: _[0]
      params:
        result_format: "'text'"

# Extracting the paper contents
- evaluate:
    paper_contents: str(_)

# Create a new persona document
- tool: create_user_doc
  arguments:
    user_id: inputs[0]['user_id']
    data:
      embed_instruction: "'Represent the query for User persona used for document retrieval: '"
      title: "'Research Papers'"
      content: _['paper_contents']