Unstructured data
The first challenge was not AI recommendation—it was preparing reliable data.
Generative AI · Networking · Automation
GenAI LinkedIn
Networking Assistant
Turning LinkedIn connections into meaningful professional conversations.
133Connections
→
5AI recommendations
→
4Responses
80%Initial response rate
Project overview
Connections do not automatically become conversations.
Professional networks can grow quickly, but having hundreds of LinkedIn connections does not necessarily lead to meaningful relationships.
I explored how Generative AI could organize connection data, identify relevant people, find time to connect, and help draft personalized outreach.
LinkedIn Data → AI Recommendations → Calendar Planning → Personalized Outreach → Real Conversations
Design question
How might Generative AI help people turn existing LinkedIn connections into meaningful professional conversations?
The goal
Organize → Match → Schedule → Connect
The assistant supports decisions while keeping the user in control of who is contacted and when.
The challenge
- Remember who is in the network
- Understand what each person does
- Identify contacts relevant to career goals
- Find time to meet
- Write personalized messages
- Follow up consistently
The data
Preparing 133 connections for analysis.
The LinkedIn export contained inconsistent formatting, repeated labels, names, titles, and dates. I used Python-assisted parsing to transform it into a reusable CSV organized by name, professional title, and connection date.
Missing contact information
Because direct outreach could not be automated, the workflow drafts LinkedIn messages for user review and manual sending.
The solution
A human-in-the-loop GenAI networking workflow.
- 01Ingest
Import LinkedIn connections.
- 02Structure
Clean and organize the data.
- 03Match
Identify contacts aligned with a networking goal.
- 04Schedule
Find potential 30-minute meeting times.
- 05Activate
Draft personalized outreach for review.
How it works
From a networking goal to personalized outreach.
Structure the network
Transform raw LinkedIn data into a clean CSV containing 133 connections.
Define a goal
The initial test focused on tech and engineering professionals in software, data, and security.
Recommend connections
The assistant selected five relevant contacts from the full network.
Find time to connect
Calendar availability suggests possible meeting times, but any calendar action requires confirmation.
Generate outreach
Each editable message combines the contact's role, the reason for connecting, and a proposed time.
Real-world testing
Four of five recommended contacts responded.
I sent personalized messages to the five connections recommended by the workflow. Four responded, producing an early 80% response rate.
This was an initial test with a small sample—not proof that the system will consistently achieve an 80% response rate.
Connections analyzed133
AI recommendations5
Outreach messages5
Responses4
Response rate80%
Meetings completedTesting in progress
Design principles
Useful AI with meaningful user control.
Human-in-the-loop
AI supports decisions but does not send messages or make consequential choices without confirmation.
Narrow & sequential
The workflow handles one task at a time: Parse → Match → Schedule → Draft → Send.
Reversible
Calendar events can be modified without restarting the recommendation process.
Personalized
Recommendations and messages respond to a specific networking goal.
Lessons learned
The workflow around AI matters as much as the output.
Data quality matters.
Reliable recommendations begin with clean, structured information.
Asking is better than assuming.
Clarifying the networking goal produced more focused recommendations.
Constraints can improve design.
Stopping at message drafting preserved user control.
AI needs real-world evaluation.
Actual outreach began measuring usefulness through responses and conversations.
Future improvements
What I am testing next.
- Semantic matchingUse embeddings or semantic similarity beyond job-title keywords.
- Networking historyTrack contacts, responses, meetings, and follow-ups.
- Response trackingCompare recommendation and message performance.
- Recurring recommendationsSuggest a small set of connections each week.
- Follow-up assistanceSupport appropriate messages after conversations.
Reflection
Building with Generative AI is not only about generating content. It requires the right data, clear constraints, user control, and evidence that the workflow produces meaningful outcomes.
The first test moved this project beyond a prototype: four of five AI-recommended connections responded, creating an initial signal to investigate through additional testing.