Sales Signals app (Agentic AI)
Automated sales coaching engine that turns B2B call transcripts into real-time, context-aware feedback, combining LLMs and historical customer data to surface revenue and…
Project Overview
My Professional Network is a local-first privacy-preserving analytics platform that transforms raw LinkedIn Takeout data into an interactive dashboard. The project demonstrates full-stack data science skills by combining ETL, NLP, unsupervised learning, and asynchronous web architecture to help users understand and strategically grow their professional networks.
Key Features & Problems Solved
Personal network intelligence from raw platform exports
Converts static archives into structured data without relying on third-party SaaS or cloud lock-in.
Semantic role clustering
Uses vector embeddings and unsupervised learning to group thousands of heterogeneous job titles into meaningful professional clusters.
Local-first, privacy-preserving
All data processing, modeling, and visualization run locally, keeping sensitive career data fully under user control.
Analysis of network structure
Reveals dominant industries, role distributions, and underrepresented areas within a personal network.
Geographic visualization
Maps global connections by resolving ambiguous location strings into precise coordinates.
Follow the link to explore the application:
Implementation Highlights
Microservices-based local architecture
Orchestrated with Docker Compose, separating web UI, background workers, database, cache, and ML inference services for scalability and isolation.
Asynchronous processing with Celery & Redis
Long-running tasks (parsing profiles, embedding generation, geocoding) execute in the background, keeping the UI responsive.
Local NLP with vector embeddings
Job titles are embedded using a locally hosted Hugging Face inference service, with aggressive caching to reduce recomputation and latency.
Unsupervised clustering pipeline
Combines dimensionality reduction (LSA / SVD) with K-Means, including a custom heuristic for automatic cluster labeling and long-tail handling.
Optimized geocoding strategy
Deduplicates and persists resolved locations to minimize API usage and control operational costs.
Read the blog post detailing the technical implementation: Building a Privacy-First LinkedIn Analytics Platform.
Outcomes & Impact
Turned opaque, static LinkedIn exports into a living analytical system.
Demonstrated how data science, backend engineering, and UX can coexist in a single cohesive product.
Showcased trade-offs around performance, privacy, scalability, and maintainability.
Created a foundation for future extensions such as graph databases and conversational network analysis.
Skills Demonstrated
Data Engineering • NLP & Embeddings • Unsupervised Learning • Network Analysis • Asynchronous Systems • Docker & DevOps • Full-Stack Data Science • Privacy-First System Design • System Architecture
Apply This to Your Business
If you have a business problem that requires data-driven solutions, feel free to reach out via contact page to discuss how I can help leverage data science, analytics, and automation to drive value for your organization.
See Also
No matching items