joseph.ergo@proton.me | Portfolio | Resume PDF | Linked-In | +212 713-617-633

Available immediately for full/part-time remote roles

METRIXLAB

Table of contents

• 1 About the project
• 2 EDA
  • 2.1 market research indestry’s yearly new recruit count
  • 2.2 metrixlab’s workforce status over the years
• 3 Persona company network graph
• 4 Workforce sample
  • 4.1 Alessandro Michelazzo
  • 4.2 Anna Dovbysh
  • 4.3 Bob Bouwense
  • 4.4 Carlie Silva
  • 4.5 Charles Pearson
  • 4.6 Chems-Eddine Makhtout
  • 4.7 Corine Makkink
  • 4.8 Craig Bloomquist
  • 4.9 Dan Kent-Smith
  • 4.10 Deb Mcdonald
  • 4.11 Ditty Menon
  • 4.12 Emanuele Carioti
  • 4.13 Frans-Anton Van Gils
  • 4.14 Hassan Mouheb
  • 4.15 Jasen Holness
  • 4.16 Kat Ferreira
  • 4.17 Kim Lijding
  • 4.18 Kjell De Orr
  • 4.19 Lizzie Nickless
  • 4.20 Marianne Bom
  • 4.21 Marjolein Van Herk-Van Tilburg
  • 4.22 Mohamed Fadl
  • 4.23 Morris Ebanks
  • 4.24 Nina Landl
  • 4.25 Robert Wilhelm
  • 4.26 Sabine Winkler
  • 4.27 Sandra Kooren
  • 4.28 Stefan De Waal
  • 4.29 Steve Becker
  • 4.30 Willem Van Muijden

## SETUP
from pathlib import Path
import duckdb
from tqdm.notebook import tqdm
import datetime
import copy
import polars as pl
import plotly.express as px
import plotly.io as pio
import re
from concurrent.futures import ThreadPoolExecutor
import plotly.graph_objects as go
import networkx as nx
import numpy as np
# pio.renderers.default = 'plotly_mimetype'
pio.renderers.default = 'jupyterlab+notebook'
pio.templates.default = "plotly_white"

path_data = Path.cwd()/'data'/'03_rdb'
path_data_companies = path_data/'companies_table.parquet'
path_data_experience = path_data/'experience_table.parquet'
path_data_emails = path_data/'emails_table.parquet'
path_data_education = path_data/'education_table.parquet'
path_data_school = path_data/'school_table.parquet'
path_data_persona = path_data/'persona_table.parquet'
path_data_profiles = path_data/'profiles_table.parquet'

path_output_images = Path.cwd()/'output'/'images'

conn = duckdb.connect()

conn.execute("SET temp_directory = 'temp';")
conn.execute("SET memory_limit = '10GB';")
conn.execute("SET max_temp_directory_size = '100GB';")
conn.execute("SET threads = 8;")
conn.execute("SET preserve_insertion_order = false;")
conn.execute("SET enable_progress_bar = true;")
conn.execute("SET enable_progress_bar_print = true;")

df = pl.read_parquet('03_target_companies3.parquet')
df_yearly_new_hires_per_indestry = pl.read_parquet('03_yearly_new_hires_per_indestry.parquet')

current_company_id = "&-friends"
current_company_id = pl.read_json("04__control__.json")[0,'current_company_id']
query = f"""
SELECT *
FROM read_parquet('{path_data_companies}')
WHERE company_id = '{current_company_id}'
"""
df_company_by_company_id = pl.DataFrame(conn.execute(query).df())

current_company_name = df_company_by_company_id[0,'company_name']
current_company_indestry = df_company_by_company_id[0,'company_industry']

current_company_parquet = Path.cwd()/'output'/'company_data'/f"{current_company_id}.parquet"

# Info about personas status from company_id

query = f"""
SELECT *
FROM read_parquet('{path_data_experience}')
WHERE company_id = '{current_company_id}'
"""
df_experiences_by_company_id = pl.DataFrame(conn.execute(query).df())

personas_whitout_end_date = df_experiences_by_company_id.filter(pl.col('end_date').is_null())
personas_who_got_raise = df_experiences_by_company_id.filter((pl.col('end_date').is_not_null()) &
                                     pl.col('persona_id').is_in(personas_whitout_end_date['persona_id'].to_list()))
personas_who_stayed = (pl
                      .concat([personas_whitout_end_date, personas_who_got_raise])
                      .sort('start_date')
                      .group_by('persona_id')
                      .agg(
                          pl.col('title_name').last(),
                          pl.col('is_primary').last(),
                          pl.col('start_date').min(),
                          pl.col('end_date').max(),
                          pl.col('title_name').count().alias('changes'),
                          pl.col('title_name').unique().alias('all_title_name'),
                      )
                      .with_columns(
                          pl.lit(True).alias('still_associated'),
                          pl.lit(None).alias('end_date')
                      )
                      .sort('changes')
                             )

personas_who_left = df_experiences_by_company_id.filter((pl.col('end_date').is_not_null()) & ~pl.col('persona_id').is_in(personas_who_stayed['persona_id'].to_list()) )
personas_who_left = (personas_who_left
                     .sort('start_date')
                     .group_by('persona_id')
                     .agg(
                          pl.col('title_name').last(),
                          pl.col('is_primary').last(),
                          pl.col('start_date').min(),
                          pl.col('end_date').max(),
                          pl.col('title_name').count().alias('changes'),
                          pl.col('title_name').unique().alias('all_title_name'),
                              )
                     .with_columns(
                         pl.lit(False).alias('still_associated'),
                         
                     )
                     .sort('changes'))

df_personas_who_worked_in_company = pl.concat([personas_who_stayed, personas_who_left], how='vertical_relaxed').with_columns(
    (pl.col('end_date').dt.year()-pl.col('start_date').dt.year()).alias('work_durration')
).sort('work_durration')

import dns.resolver
import smtplib
import socket

def check_deliverability(email_address):
    """
    Checks the deliverability of an email address by verifying MX records
    and performing an SMTP connection test.
    """
    if '@' not in email_address:
        return False
    
    domain = email_address.split('@')[1]
    
    # Check for MX records
    try:
        mx_records = dns.resolver.resolve(domain, 'MX')
        if not mx_records:
            return False
    except (dns.resolver.NoAnswer, dns.resolver.NXDOMAIN, dns.resolver.Timeout):
        return False

    # Perform SMTP connection test
    mx_host = str(mx_records[0].exchange)
    
    # Validate MX hostname before attempting connection
    try:
        # Test if hostname can be properly encoded
        mx_host.encode('idna')
    except UnicodeError:
        return False
    
    try:
        with smtplib.SMTP(mx_host, timeout=10) as smtp:
            smtp.set_debuglevel(0)
            smtp.helo(socket.gethostname())
            smtp.mail('test@example.com')
            code, _ = smtp.rcpt(email_address)

            return code == 250  # 250 indicates valid email address
            
    except (smtplib.SMTPException, socket.error, UnicodeError):
        return False

# info of all personas info
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_persona}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas = pl.DataFrame(conn.execute(query).df())

# info of all personas profiles
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_profiles}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_profile = pl.DataFrame(conn.execute(query).df())
df_all_personas_profile_f = df_all_personas_profile.group_by('persona_id').agg(pl.col('url').unique())

# info of all personas email
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")
list_for_in = ', '.join(list_w)

query = f"""
SELECT *
FROM read_parquet('{path_data_emails}')
WHERE persona_id IN ({list_for_in}) AND type == 'personal'
"""
df_all_personas_emails = pl.DataFrame(conn.execute(query).df())

def def_polars_fix_gmail(x):
    if "@gmail" in x:
        first_part = x.split('@')[0]
        second_part = x.split('@')[1]
        return f"{first_part.replace(".",'')}@{second_part}"
    else:
        return x

df_all_personas_emails_f = (df_all_personas_emails
                            .with_columns(pl.col('address')
                                          .map_elements(def_polars_fix_gmail, return_dtype=pl.String)
                                          .alias('normalised_emails'))
                            .unique('normalised_emails', keep='first')
                            .sort('persona_id')
                            .drop('normalised_emails')
                         )
df_all_personas_emails_f = (df_all_personas_emails_f.group_by('persona_id').agg(pl.col('address').unique(),pl.col('type').unique()))
df_all_personas_plus = df_all_personas.join(df_all_personas_emails_f, on='persona_id', how='left')

df_full_personas_who_worked_in_company = (df_personas_who_worked_in_company
                                       .join(df_all_personas_plus, on='persona_id', how='left')
                                       .join(df_all_personas_profile_f, on='persona_id', how='left')
                                      )

df_full_personas_who_worked_in_company = (
    df_full_personas_who_worked_in_company.with_columns(
        (pl.col("start_date").fill_null(pl.col("start_date").min()))
        .dt.year()
        .alias("start_year"),
        (pl.col("end_date").dt.year()).alias("end_year"),
    )
)

work_years = []
for i in range(len(df_full_personas_who_worked_in_company)):
    start_y = df_full_personas_who_worked_in_company[i, "start_year"]
    if df_full_personas_who_worked_in_company[i, "end_year"]:
        end_y = df_full_personas_who_worked_in_company[i, "end_year"]
    else:
        end_y = 2020

    tmp_work_years = []
    for y in range(start_y, end_y + 1):
        tmp_work_years.append(y)

    work_years.append(tmp_work_years)

df_full_personas_who_worked_in_company = (
    df_full_personas_who_worked_in_company.with_columns(
        pl.Series("work_years", work_years)
    )
)

# add hireups
title_name_match = ["ceo","chief","founder","owner","president","vp","vice","director",
    "cfo","cto","partner","head of","hr ","human","talent","senior","manager","lead"]

df_full_personas_who_worked_in_company = (df_full_personas_who_worked_in_company
    .with_columns(
        pl.when(pl.col('title_name').str.contains_any(title_name_match)).then(True).otherwise(False).alias("higher_up")
    ))



df_tmp_email_checker = (
    df_full_personas_who_worked_in_company
    .filter(
            pl.col('still_associated')==True,
            pl.col('address').list.len()>0
    )
        ['persona_id','address']
        .explode('address')
)

# if current_company_parquet.exists():
#     df_pre_full_personas_who_worked_in_company = pl.read_parquet(current_company_parquet)
#     list_pre_deliverable_address = df_pre_full_personas_who_worked_in_company['address'].drop_nulls().explode().to_list()
# else:
#     list_pre_deliverable_address = []

# list_of_emails_to_check = df_tmp_email_checker['address'].drop_nulls().to_list()
# list_lists_email_check = []

# var_total_emails = len(list_of_emails_to_check)
# var_current_email_count = 0

# def def_check_and_populate(email_to_check):
#     global list_lists_email_check, var_current_email_count
#     if email_to_check in list_pre_deliverable_address:
#         list_lists_email_check.append([email_to_check, True])
#     elif '@gmail' in email_to_check:
#         list_lists_email_check.append([email_to_check, True])
#     else:
#         try:
#             is_deliverable = check_deliverability(email_to_check)
#             list_lists_email_check.append([email_to_check, is_deliverable])
#         except:
#             list_lists_email_check.append([email_to_check, False])
#     var_current_email_count += 1
#     print(' '*10, end='\r')
#     print(round(var_current_email_count/var_total_emails,5), end='\r')

# with ThreadPoolExecutor(max_workers=20) as executor:
#     results = list(executor.map(def_check_and_populate, list_of_emails_to_check))

# df_email_check = pl.DataFrame(list_lists_email_check, schema=["address", "deliverable"], orient="row")
# try:
#     df_tmp_email_checker_f = (
#         df_tmp_email_checker
#             .join(df_email_check, on='address')
#             .filter(pl.col('deliverable')==True)
#             .group_by('persona_id').agg(pl.col('address').unique().alias("deliverable_address"))
#     )
# except:
#     df_tmp_email_checker_f = pl.DataFrame()

# if df_tmp_email_checker_f.is_empty():
#     df_full_personas_who_worked_in_company = df_full_personas_who_worked_in_company.join(df_tmp_email_checker.rename({'address':'deliverable_address'}), on="persona_id", how='left')
# else:
#     df_full_personas_who_worked_in_company = df_full_personas_who_worked_in_company.join(df_tmp_email_checker_f, on="persona_id", how='left')

# Info about personas experiences

# info of all experiences[]
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT *
FROM read_parquet('{path_data_experience}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_experiences = pl.DataFrame(conn.execute(query).df())


# info of all comapnies in said experiences
list_w = []
for word in df_all_personas_experiences['company_id'].unique().to_list():
    if "'" not in word:
        list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT company_id, company_name, company_industry, company_linkedin_url, company_location_country
FROM read_parquet('{path_data_companies}')
WHERE company_id IN ({list_for_in})
"""
df_all_companies = pl.DataFrame(conn.execute(query).df())

df_full_personas_experiences_plus = df_all_personas_experiences.join(df_all_companies, on='company_id', how='left')

df_full_personas_experiences_plus = (
    df_full_personas_experiences_plus
    .with_columns(
        pl.when(
            pl.col('company_id')==current_company_id
        )
        .then(True)
        .otherwise(False)
        .alias('target')
    )
)

# Info about personas education

# info of all experiences
list_w = []
for word in df_experiences_by_company_id['persona_id'].unique().to_list():
    list_w.append(f"'{word}'")

list_for_in = ', '.join(list_w)
query = f"""
SELECT *
FROM read_parquet('{path_data_education}')
WHERE persona_id IN ({list_for_in})
"""
df_all_personas_education = pl.DataFrame(conn.execute(query).df())


#ifon of allcomapnies in said experiences
list_w = []
for word in df_all_personas_education['school_id'].unique().to_list():
    if "'" not in word:
        list_w.append(f"'{word}'")

if list_w:
    list_for_in = ', '.join(list_w)
    query = f"""
    SELECT school_id, school_name, school_type, school_website, school_location_country
    FROM read_parquet('{path_data_school}')
    WHERE school_id IN ({list_for_in})
    """
    df_all_school = pl.DataFrame(conn.execute(query).df())
    
    df_full_personas_education_plus = df_all_personas_education.join(df_all_school, on='school_id', how='left')
else:
    df_full_personas_education_plus = df_all_personas_education

1 About the project

The project came to life after realizing that web scraping doesn’t allow deep-level filtering—without consuming too much time.The irony is, this project itself took me about a month, but the final RDB contains more data than I could ever scrape.

The raw data was 1.4 TB in size and holds information previously scraped.
Processing was done on my local machine using Python, Polars, and DuckDB, following this workflow:
- Processed raw data into structured Parquet files using Polars.
- Transformed each Parquet file into mini RDBs using Polars.
- Merged all mini RDBs into one using DuckDB.
- Analyzed and filtered data to fit the current project.

2 EDA

2.1 market research indestry’s yearly new recruit count

list_of_unique_company_experience_years = []
for y in df_full_personas_who_worked_in_company['start_year'].unique().drop_nulls().to_list():
    if y not in list_of_unique_company_experience_years:
        list_of_unique_company_experience_years.append(y)
for y in df_full_personas_who_worked_in_company['end_year'].unique().drop_nulls().to_list():
    if y not in list_of_unique_company_experience_years:
        list_of_unique_company_experience_years.append(y)

list_year = []
list_state = []
list_count = []
list_names = []

def def_get_names_breked(tmp):
    if tmp.is_empty():
        names_string = ''
    else:
        tmp_list_name = []
        names_limit = 3
        row_limit = names_limit * 6
        for i, name in enumerate(tmp['full_name'].to_list()):
            ii = i+1
            tmp_list_name.append(name.title())
            if ii!=0 and ii%names_limit==0:
                tmp_list_name.append("<br>")
            if ii==row_limit:
                tmp_list_name.append("...")
                break
        names_string = ', '.join(tmp_list_name).replace(", <br>, ","<br>")
    return names_string

for y in list_of_unique_company_experience_years:
    #recuite state
    list_year.append(y)
    list_state.append('Recruited')
    tmp = df_full_personas_who_worked_in_company.filter(pl.col('start_year')==y).sort('full_name')
    list_count.append(len(tmp))
    list_names.append(def_get_names_breked(tmp))
    
    #recuite state
    list_year.append(y)
    list_state.append('Resigned')
    tmp = df_full_personas_who_worked_in_company.filter(pl.col('end_year')==y).sort('full_name')
    list_count.append(len(tmp))
    list_names.append(def_get_names_breked(tmp))

df_m_recruite_vs_resign = pl.DataFrame({
    'year':list_year,
    'status':list_state,
    'count':list_count,
    'names':list_names,})

2.2 metrixlab’s workforce status over the years

3 Persona company network graph

gr_net = df_full_personas_experiences_plus.with_columns(pl.col('company_id').str.to_uppercase()).group_by('persona_id','company_id').agg(pl.len().alias('count')).sort('count')
list_top_in_network = gr_net['company_id'].value_counts().sort('count', descending=True)['company_id'].to_list()[:5]
gr_net_f = gr_net.filter(pl.col('company_id').is_in(list_top_in_network))

list_letters = ['A','B','C','D','E','F','G','H']
dict_company = {}
dict_company_rev = {}
for company, letter in zip(list_top_in_network, list_letters ):
    dict_company[letter] = company
    dict_company_rev[company] = letter

gr_gr_net_f = gr_net_f.sort('company_id').group_by('persona_id').agg(pl.col('company_id').unique().sort(),)

gr_gr_net_f2 = (
    gr_gr_net_f['company_id']
    .value_counts()
    .with_columns(
        # pl.col('company_id').list.join(', '),
        (pl.col('count')/len(gr_gr_net_f)).alias('per')
    )
    .sort('per',descending=True)
)

list_prob = []
for i in range(len(gr_gr_net_f2)):
    tmp_prob_letters = []
    for k in dict_company.keys():
        if dict_company[k] in gr_gr_net_f2[i]['company_id'][0].to_list():
            tmp_prob_letters.append(f' {k}')
        else:
            tmp_prob_letters.append(f'¬{k}')

    list_prob.append(f"P({' ∩ '.join(tmp_prob_letters)}) = {round(gr_gr_net_f2[i]['per'][0],4)}")
annon_prob_text = "<b>Probability Distribution:</b><br>" + '<br>'.join(list_prob)



# Create network graph
G = nx.Graph()
for persona, company in gr_net_f.select(['persona_id', 'company_id']).iter_rows():
    G.add_edge(persona, company)

# Get unique values
persona_ids = gr_net_f['persona_id'].unique().to_list()
company_ids = gr_net_f['company_id'].unique().to_list()

# Calculate degrees (connection counts)
degree_dict = dict(G.degree())

# Get min and max degrees for scaling
company_degrees = [degree_dict[c] for c in company_ids]
persona_degrees = [degree_dict[p] for p in persona_ids]

min_company_degree = min(company_degrees) if company_degrees else 1
max_company_degree = max(company_degrees) if company_degrees else 1
min_persona_degree = min(persona_degrees) if persona_degrees else 1
max_persona_degree = max(persona_degrees) if persona_degrees else 1

# Define size ranges
COMPANY_MIN_SIZE = 25
COMPANY_MAX_SIZE = 100
PERSONA_MIN_SIZE = 5
PERSONA_MAX_SIZE = 20

# print(f"Company connections range: {min_company_degree} - {max_company_degree}")
# print(f"Persona connections range: {min_persona_degree} - {max_persona_degree}")

# Sort companies by degree (size) in descending order
company_ids_sorted = sorted(company_ids, key=lambda x: degree_dict[x], reverse=True)

# Check if "Nokia" exists in the data
HIGHLIGHTED_COMPANY = current_company_id
HIGHLIGHTED_COMPANY_EXISTS = HIGHLIGHTED_COMPANY.lower() in [str(c).lower() for c in company_ids]

if HIGHLIGHTED_COMPANY_EXISTS:
    # Get the actual case-sensitive name
    highlighted_company = next(c for c in company_ids if str(c).lower() == HIGHLIGHTED_COMPANY.lower())
    # print(f"Highlighting company: {highlighted_company} (with {degree_dict[highlighted_company]} connections)")
else:
    # print(f"Warning: '{HIGHLIGHTED_COMPANY}' not found in company list")
    highlighted_company = None

# Create layout (companies on outer circle, ordered by size)
pos = {}
num_companies = len(company_ids_sorted)
radius_outer = 2.0

# Position companies on circle, ordered by size (largest first)
for i, company in enumerate(company_ids_sorted):
    # Start at top (90° or π/2 radians) and go counter-clockwise (add angle)
    # Counter-clockwise rotation: angle = start_angle + (i * 2π / num_companies)
    # This puts largest at top, next on left, then bottom, then right
    start_angle = np.pi / 2  # 90° at top
    
    # For counter-clockwise rotation
    angle = start_angle - (2 * np.pi * i / num_companies)
    
    # Convert to x, y coordinates
    pos[company] = (radius_outer * np.cos(angle), radius_outer * np.sin(angle))

# Position personas
for i, persona in enumerate(persona_ids):
    connected_companies = [c for c in company_ids if G.has_edge(persona, c)]
    if connected_companies:
        avg_x = np.mean([pos[c][0] for c in connected_companies])
        avg_y = np.mean([pos[c][1] for c in connected_companies])
        # Add jitter to spread out personas
        jitter_x = np.random.uniform(-0.2, 0.2)
        jitter_y = np.random.uniform(-0.2, 0.2)
        pos[persona] = (avg_x * 0.5 + jitter_x, avg_y * 0.5 + jitter_y)
    else:
        pos[persona] = (0, 0)

# Prepare edge traces
edge_x, edge_y = [], []
for edge in G.edges():
    x0, y0 = pos[edge[0]]
    x1, y1 = pos[edge[1]]
    edge_x.extend([x0, x1, None])
    edge_y.extend([y0, y1, None])

edge_trace = go.Scatter(
    x=edge_x, y=edge_y,
    line=dict(width=0.6, color='rgba(120, 120, 120, 0.15)'),
    hoverinfo='none',
    mode='lines')

# Prepare node traces with proportional sizing
company_x, company_y, company_text = [], [], []
company_color, company_size, company_hover = [], [], []
company_border_width = []  # For border thickness
company_border_color = []  # For border color

persona_x, persona_y = [], []
persona_color, persona_size, persona_hover = [], [], []

# Helper function to scale size proportionally
def scale_size(value, min_val, max_val, min_size, max_size):
    if max_val == min_val:
        return (min_size + max_size) / 2
    return min_size + (value - min_val) / (max_val - min_val) * (max_size - min_size)

# Add COMPANY nodes in sorted order (largest first)
for company in company_ids_sorted:
    x, y = pos[company]
    company_x.append(x)
    company_y.append(y)
    company_text.append(str(company))
    company_color.append('#EF553B')
    
    connections = degree_dict[company]
    # Scale size based on connection count
    scaled_size = scale_size(
        connections, 
        min_company_degree, 
        max_company_degree,
        COMPANY_MIN_SIZE, 
        COMPANY_MAX_SIZE
    )
    company_size.append(scaled_size)
    
    # Custom border for highlighted company
    if highlighted_company and company == highlighted_company:
        company_border_width.append(4)  # Thicker border
        company_border_color.append('#000000')  # Black border
    else:
        company_border_width.append(1)
        company_border_color.append('#000000')
    
    # Hover text
    personas = gr_net_f.filter(pl.col('company_id') == company)['persona_id'].to_list()
    rank = company_ids_sorted.index(company) + 1
    hover_text = f"<b>Company #{rank}:</b> {company}<br>"
    hover_text += f"<b>Personas worked here:</b> {connections}<br>"
    hover_text += f"<b>Connection rank:</b> {rank}/{len(company_ids_sorted)}<br>"
    if connections > 0:
        for persona in personas[:5]:
            persona_name = df_all_personas.filter(pl.col('persona_id')==persona)['full_name'][0].title()
            hover_text += f" • {persona_name}<br>"
        if connections > 5:
            hover_text += f" • ... and {connections - 5} more"
    company_hover.append(hover_text)

# Add PERSONA nodes
for persona in persona_ids:
    x, y = pos[persona]
    persona_x.append(x)
    persona_y.append(y)
    persona_color.append('#636efa')
    
    connections = degree_dict[persona]
    # Scale size based on connection count
    scaled_size = scale_size(
        connections,
        min_persona_degree,
        max_persona_degree,
        PERSONA_MIN_SIZE,
        PERSONA_MAX_SIZE
    )
    persona_size.append(scaled_size)
    
    # Hover text
    companies = gr_net_f.filter(pl.col('persona_id') == persona)['company_id'].to_list()
    persona_name = df_all_personas.filter(pl.col('persona_id')==persona)['full_name'][0].title()
    hover_text = f"<b>Persona:</b> {persona_name}<br>"
    hover_text += f"<b>Companies worked at:</b> {connections}<br>"
    if connections > 0:
        # Check if worked at highlighted company
        if highlighted_company:
            worked_at_highlighted = highlighted_company in companies
            if worked_at_highlighted:
                hover_text += f"<b>Worked at {highlighted_company}:</b> ✓<br>"
        
        hover_text += "<br>".join([f"  • {comp}" for comp in companies[:5]])
        if connections > 5:
            hover_text += f"<br>  • ... and {connections - 5} more"
    persona_hover.append(hover_text)

# Create company node trace
company_trace = go.Scatter(
    x=company_x, y=company_y,
    mode='markers+text',
    hoverinfo='text',
    hovertext=company_hover,
    text=company_text,
    textposition="top center",
    textfont=dict(size=14, color='black'),
    marker=dict(
        color=company_color,
        size=company_size,
        line=dict(
            width=company_border_width,
            color=company_border_color
        ),
        opacity=0.9)
)

# Create persona node trace
persona_trace = go.Scatter(
    x=persona_x, y=persona_y,
    mode='markers',
    hoverinfo='text',
    hovertext=persona_hover,
    text=None,  # No text for personas
    marker=dict(
        color=persona_color,
        size=persona_size,
        line=dict(width=1, color='black'),
        opacity=0.7)
)

# Calculate axis ranges for 1:1 aspect ratio
all_positions = list(pos.values())
x_vals = [p[0] for p in all_positions]
y_vals = [p[1] for p in all_positions]

# Add padding
x_range = [min(x_vals) - 0.5, max(x_vals) + 0.5]
y_range = [min(y_vals) - 0.5, max(y_vals) + 0.5]

# Make axes have the same range for 1:1 aspect
max_range = max(x_range[1] - x_range[0], y_range[1] - y_range[0])
x_center = (x_range[0] + x_range[1]) / 2
y_center = (y_range[0] + y_range[1]) / 2

x_range = [x_center - max_range/2, x_center + max_range/2]
y_range = [y_center - max_range/2, y_center + max_range/2]

# Create figure with 1:1 aspect ratio
fig = go.Figure(data=[edge_trace, persona_trace, company_trace],
                layout=go.Layout(
                    title=f'Persona-Company Network (Companies Ordered by Size)<br><sup>Highlighted: {highlighted_company if highlighted_company else "None"}</sup>',
                    showlegend=False,
                    hovermode='closest',
                    margin=dict(b=20, l=20, r=20, t=100),
                    xaxis=dict(
                        showgrid=False, 
                        zeroline=False, 
                        showticklabels=False,
                        range=x_range,
                        scaleanchor="y",
                        scaleratio=1
                    ),
                    yaxis=dict(
                        showgrid=False, 
                        zeroline=False, 
                        showticklabels=False,
                        range=y_range
                    ),
                    plot_bgcolor='white',
                    paper_bgcolor='white',
                    width=900,
                    height=900
                ))

# Add legend with size examples and highlighting info
# legend_text = f"""
# <b>Node Size = Connection Count</b><br>
# <span style='color:#EF553B'>● Companies</span><br>
# <span style='color:#636efa'>● Personas</span> (hover for details)
# """

# fig.add_annotation(
#     x=0.98, y=0.98,
#     xref="paper", yref="paper",
#     text=legend_text,
#     showarrow=False,
#     font=dict(size=14),
#     align="left",
#     bgcolor="rgba(255, 255, 255, 0.95)",
    
# )

# Add top companies list
top_companies = company_ids_sorted[:10]  # Top 10 companies
top_companies_text = "<b>Top Companies by Connections:</b><br>"
for i, company in enumerate(top_companies, 1):
    connections = degree_dict[company]
    top_connections = degree_dict[top_companies[0]]
    connections_per = f" | {round(connections/top_connections*100)}%" if highlighted_company and company != highlighted_company else ""
    highlight_indicator = " " if highlighted_company and company == highlighted_company else ""
    top_companies_text += f"{dict_company_rev[company]}. {company}: {connections} {connections_per} {highlight_indicator}<br>"

fig.add_annotation(
    x=0.02, y=0.98,
    xref="paper", yref="paper",
    text=top_companies_text,
    showarrow=False,
    font=dict(size=14),
    align="left",
    bgcolor="rgba(255, 255, 255, 0.9)",
    # bordercolor="#666",
    # borderwidth=1
)

# Add probabiliy list

fig.add_annotation(
    x=0.98, y=0.98,
    xref="paper", yref="paper",
    text=annon_prob_text,
    showarrow=False,
    font=dict(
        family="'Courier New', monospace",  # Multiple fallbacks
        size=12,
        color="black"
    ),
    align="left",
    bgcolor="rgba(255, 255, 255, 0.95)",
    
)
fig.write_image((path_output_images/f'network_{current_company_id}.webp'))
fig.show()

amount = 5

tmp = df_full_personas_who_worked_in_company.sort(
    ["inferred_salary", "linkedin_connections", "inferred_years_experience"],
    descending=True,
)
tmp_gr = df_full_personas_experiences_plus.group_by('persona_id').agg(pl.len().alias('experience_count'))
tmp = df_full_personas_who_worked_in_company.join(tmp_gr, on='persona_id').sort('experience_count',descending=True)

tmp2 = pl.concat(
    [tmp.filter(pl.col('still_associated')==True, pl.col('higher_up')==True)[:amount*2],
     tmp.filter(pl.col('still_associated')==True, pl.col('higher_up')==False)[:amount*2],
     tmp.filter(pl.col('still_associated')==False, pl.col('higher_up')==True)[:amount*1],
     tmp.filter(pl.col('still_associated')==False, pl.col('higher_up')==False)[:amount*1],
    ]
).sort("full_name")

list_persona_for_plot = tmp2['persona_id'].to_list()

# Workforce data

def def_plotly_experience_range(current_persona_id):
    tmp_df = (df_full_personas_who_worked_in_company
              .filter(pl.col('persona_id')==current_persona_id)
              .with_columns(pl.col('end_year').fill_null(2021))['start_year','end_year'])
    
    fig_tmp = copy.deepcopy(fig_company_hiring_trend)
    fig_tmp.add_vrect(
        x0=tmp_df[0,'start_year'],
        x1=tmp_df[0,'end_year'],
        fillcolor="blue",
        opacity=0.1,
        line_width=0 
    )
    return fig_tmp

def def_plotly_experience_gantt(current_persona_id):
    px_data = (df_full_personas_experiences_plus
               .filter(pl.col('persona_id')==current_persona_id)
               .with_columns(
                   pl.col('end_date').fill_null(datetime.datetime(2020, 1, 1, 0,0)),
                   pl.col('company_name').str.to_uppercase(),
                   # pl.col('company_name').str.to_uppercase().str.replace_all('&', '-and-')
               )
               .sort('start_date'))
    
    y_order = px_data['company_name'].to_list()
    
    fig = px.timeline(px_data,x_start="start_date", x_end="end_date", y="company_name",
                      color='target',hover_data=["title_name"], height=140+30*len(px_data),
                      category_orders={"company_name": y_order},
                      color_discrete_map={True:'#EF553B',  False:'#636efa'},
                      labels={'target':'Target', 'start_date':'Recruited', 'end_date':'If-Resigned', 
                             'company_name':'Company', 'title_name':'Job role'}
                     # title=f"Experience of {current_persona_name}.",
                     )
    fig.update_yaxes(
        # autorange="reversed",
                              showgrid=True,
                              gridcolor='lightgray',
                              gridwidth=1,
                              griddash='dot'
    )
    fig.update_layout(showlegend=False, xaxis_title=None, yaxis_title=None)
    return fig

4 Workforce sample

4.1 Alessandro Michelazzo

Job title: Visual designer
Socials: https://linkedin.com/in/alemichelazzo | https://twitter.com/alemichelazzo | https://linkedin.com/in/alessandromichelazzo

4.1.1 Alessandro Michelazzo’s working period at metrixlab

4.1.2 Gantt plot of Alessandro Michelazzo’s experience

---

4.2 Anna Dovbysh

Job title: Social media analyst
Socials: https://linkedin.com/in/anna-dovbysh-18b42728 | https://facebook.com/dovbysh.anna

4.2.1 Anna Dovbysh’s working period at metrixlab

4.2.2 Gantt plot of Anna Dovbysh’s experience

---

4.3 Bob Bouwense

Job title: Freelance photo editor
Socials: https://linkedin.com/in/bobbouwense

4.3.1 Bob Bouwense’s working period at metrixlab

4.3.2 Gantt plot of Bob Bouwense’s experience

---

4.4 Carlie Silva

Job title: Qualitative research analyst
Socials: https://linkedin.com/in/carlie-silva-4898aa100 | https://facebook.com/carlie.silva

4.4.1 Carlie Silva’s working period at metrixlab

4.4.2 Gantt plot of Carlie Silva’s experience

---

4.5 Charles Pearson

Job title: Senior vice president sales
Socials: https://linkedin.com/in/charles-pearson-261363 | https://linkedin.com/in/charlespearson | https://twitter.com/cpmktresearch

4.5.1 Charles Pearson’s working period at metrixlab

4.5.2 Gantt plot of Charles Pearson’s experience

---

4.6 Chems-Eddine Makhtout

Job title: Financial controller - ad int
Socials: https://linkedin.com/in/chemseddine

4.6.1 Chems-Eddine Makhtout’s working period at metrixlab

4.6.2 Gantt plot of Chems-Eddine Makhtout’s experience

---

4.7 Corine Makkink

Job title: Client director
Socials: https://linkedin.com/in/corine-makkink-156bb24 | https://linkedin.com/in/corinemakkink

4.7.1 Corine Makkink’s working period at metrixlab

4.7.2 Gantt plot of Corine Makkink’s experience

---

4.8 Craig Bloomquist

Job title: Srm
Socials: https://linkedin.com/in/craig-bloomquist-362916 | https://facebook.com/craig.bloomquist.9

4.8.1 Craig Bloomquist’s working period at metrixlab

4.8.2 Gantt plot of Craig Bloomquist’s experience

---

4.9 Dan Kent-Smith

Job title: Client director
Socials: https://linkedin.com/in/dan-kent-smith-a8352512 | https://linkedin.com/in/dankentsmith

4.9.1 Dan Kent-Smith’s working period at metrixlab

4.9.2 Gantt plot of Dan Kent-Smith’s experience

---

4.10 Deb Mcdonald

Job title: Consultant - online community moderator
Socials: https://linkedin.com/in/deb-mcdonald-876441 | https://twitter.com/debmcdsf

4.10.1 Deb Mcdonald’s working period at metrixlab

4.10.2 Gantt plot of Deb Mcdonald’s experience

---

4.11 Ditty Menon

Job title: Big data consultant ai for the data artists
Socials: https://linkedin.com/in/dittymenon | https://nl.linkedin.com/in/dittymenon

4.11.1 Ditty Menon’s working period at metrixlab

4.11.2 Gantt plot of Ditty Menon’s experience

---

4.12 Emanuele Carioti

Job title: Opinionista opionionbar.com - metrixlab nederland b v
Socials: https://linkedin.com/in/emanuele-carioti-61333718 | https://facebook.com/emanuelecarioti | https://twitter.com/emanuelecarioti

4.12.1 Emanuele Carioti’s working period at metrixlab

4.12.2 Gantt plot of Emanuele Carioti’s experience

---

4.13 Frans-Anton Van Gils

Job title: Product owner
Socials: https://meetup.com/members/7593415 | https://quora.com/frans-anton-van-gils | https://facebook.com/fransvgls | https://github.com/fvangils | https://twitter.com/fvangils | https://linkedin.com/in/vangils

4.13.1 Frans-Anton Van Gils’s working period at metrixlab

4.13.2 Gantt plot of Frans-Anton Van Gils’s experience

---

4.14 Hassan Mouheb

Job title: Managing director
Socials: https://linkedin.com/in/hassan-mouheb-12bb88

4.14.1 Hassan Mouheb’s working period at metrixlab

4.14.2 Gantt plot of Hassan Mouheb’s experience

---

4.15 Jasen Holness

Job title: Senior vice president - agency and media practice
Socials: https://linkedin.com/in/jasen-holness-47944a28

4.15.1 Jasen Holness’s working period at metrixlab

4.15.2 Gantt plot of Jasen Holness’s experience

---

4.16 Kat Ferreira

Job title: Marketing manager, zoomerang, markettools inc
Socials: https://linkedin.com/in/katferreira | https://linkedin.com/in/kathy-ferreira-4593983 | https://linkedin.com/in/kathyferreira

4.16.1 Kat Ferreira’s working period at metrixlab

4.16.2 Gantt plot of Kat Ferreira’s experience

---

4.17 Kim Lijding

Job title: Junior research manager big data analytics
Socials: https://linkedin.com/in/kim-lijding-46380365 | https://facebook.com/kim.lijding | https://linkedin.com/in/kimlijding

4.17.1 Kim Lijding’s working period at metrixlab

4.17.2 Gantt plot of Kim Lijding’s experience

---

4.18 Kjell De Orr

Job title: Group commercial board director
Socials: https://linkedin.com/in/kjell-de-orr-270a17 | https://facebook.com/kjell.deorr

4.18.1 Kjell De Orr’s working period at metrixlab

4.18.2 Gantt plot of Kjell De Orr’s experience

---

4.19 Lizzie Nickless

Job title: Global marketing specialist
Socials: https://linkedin.com/in/lizzie-nickless | https://linkedin.com/in/lizzie-nickless-6aab366b

4.19.1 Lizzie Nickless’s working period at metrixlab

4.19.2 Gantt plot of Lizzie Nickless’s experience

---

4.20 Marianne Bom

Job title: Research manager new product development and innovation
Socials: https://linkedin.com/in/mariannebom

4.20.1 Marianne Bom’s working period at metrixlab

4.20.2 Gantt plot of Marianne Bom’s experience

---

4.21 Marjolein Van Herk-Van Tilburg

Job title: Global general counsel and dpo
Socials: https://linkedin.com/in/marjoleinvantilburg | https://twitter.com/marjovhvt

4.21.1 Marjolein Van Herk-Van Tilburg’s working period at metrixlab

4.21.2 Gantt plot of Marjolein Van Herk-Van Tilburg’s experience

---

4.22 Mohamed Fadl

Job title: Social media analyst
Socials: https://twitter.com/m_fadl | https://linkedin.com/in/mfadl

4.22.1 Mohamed Fadl’s working period at metrixlab

4.22.2 Gantt plot of Mohamed Fadl’s experience

---

4.23 Morris Ebanks

Job title: Research manager
Socials: https://facebook.com/mrswabyebanks | https://linkedin.com/in/mswabyebanks

4.23.1 Morris Ebanks’s working period at metrixlab

4.23.2 Gantt plot of Morris Ebanks’s experience

---

4.24 Nina Landl

Job title: Digital and global marketing intern
Socials: https://linkedin.com/in/nina-landl-397a89104

4.24.1 Nina Landl’s working period at metrixlab

4.24.2 Gantt plot of Nina Landl’s experience

---

4.25 Robert Wilhelm

Job title: Board member and investor
Socials: https://linkedin.com/in/robert-wilhelm-b01107 | https://facebook.com/robert.wilhelm.nl | https://linkedin.com/in/robertwilhelm | https://twitter.com/rpwilhelm

4.25.1 Robert Wilhelm’s working period at metrixlab

4.25.2 Gantt plot of Robert Wilhelm’s experience

---

4.26 Sabine Winkler

Job title: Global marketing and communications director
Socials: https://twitter.com/sabiine | https://linkedin.com/in/sabine-winkler-6a36b5a | https://linkedin.com/in/sabinewinkler

4.26.1 Sabine Winkler’s working period at metrixlab

4.26.2 Gantt plot of Sabine Winkler’s experience

---

4.27 Sandra Kooren

Job title: Senior research manager
Socials: https://linkedin.com/in/sandra-kooren-01483a3 | https://facebook.com/sandra.kooren.9

4.27.1 Sandra Kooren’s working period at metrixlab

4.27.2 Gantt plot of Sandra Kooren’s experience

---

4.28 Stefan De Waal

Job title: Senior research manager
Socials: https://linkedin.com/in/stefan-de-waal-73527714 | https://linkedin.com/in/stefandewaal

4.28.1 Stefan De Waal’s working period at metrixlab

4.28.2 Gantt plot of Stefan De Waal’s experience

---

4.29 Steve Becker

Job title: Vp, client director
Socials: https://linkedin.com/in/steve-becker-7302549 | https://facebook.com/steve.becker.39501

4.29.1 Steve Becker’s working period at metrixlab

4.29.2 Gantt plot of Steve Becker’s experience

---

4.30 Willem Van Muijden

Job title: Designer
Socials: https://linkedin.com/in/willem-vrederick-van-muijden-5a05071b

4.30.1 Willem Van Muijden’s working period at metrixlab

4.30.2 Gantt plot of Willem Van Muijden’s experience

---

df_full_personas_who_worked_in_company.write_parquet(current_company_parquet)