GradJobsUK

GradJobsUK—a real-time UK graduate job board tailored for international CS/Data students. Auto-scrapes LinkedIn every 6 hours, filters visa-sponsored roles, and matches your CV to jobs.

Comment

GradJobsUK

Inspiration

Every graduate student faces the overwhelming pressure of navigating hundreds of unstructured LinkedIn job descriptions. For international students, there's an additional, critical pain point: identifying which companies actually offer visa sponsorship.

Our research revealed two core problems with traditional platforms:

1. Unstructured Data

Job descriptions are massive blocks of text, making bulk filtering nearly impossible.

2. Unreliable Visa Information

Companies with sponsor licenses often omit this information, while others include the word "sponsor" only to state "this role does not offer sponsorship."
Manually screening these takes 1–3 hours daily.

To solve this, we applied Text Mining and Correlation Analysis to build GradJobsUK — a quantified, scalable, and highly relevant data platform tailored specifically for graduate students in the tech space.

What It Does

GradJobsUK is a job aggregation platform anchored on the job seeker's specific skills.

It automatically scrapes LinkedIn every 6 hours, exclusively targeting CS and Data roles:

  • Software Engineering (SWE)
  • Machine Learning (ML)
  • DevOps
  • Quant
  • Data-related roles

The system only collects jobs that:

  • Were posted within the last 24 hours
  • Have fewer than 100 applicants

User Experience

The platform focuses on high-efficiency data consumption.

Interactive Dashboard

Features:

  • 4 macro statistical metrics
  • 3 visual charts

All components dynamically sync with a global time filter, allowing users to quickly identify market trends.

Advanced Filtering

Users can rapidly narrow down opportunities using:

  • Top search bar
  • Left-side multi-variable filters

Excel-Grade Job List

Results are displayed in a virtualized table that supports:

  • Free sorting
  • Column filtering
  • Direct links to the original LinkedIn job post
  • One-click Excel export

CV Matching

Users can upload their CV.

The system then:

  1. Calculates a TF-IDF cosine similarity score
  2. Compares the CV with all active job listings
  3. Adds the match percentage as a sortable column

International Ready

The platform supports 5 languages:

  • English (EN)
  • Chinese (ZH)
  • French (FR)
  • Spanish (ES)
  • Dutch (NL)

It also includes our core 5-tier visa status verification system.

How I Built It (Under the Hood)

The system was engineered to address two major technical challenges.

1. Caching Architecture (Read/Write Separation)

Users never trigger LinkedIn requests directly.

Write Path

  • APScheduler triggers background tasks
  • Python FastAPI (async) scrapers run periodically
  • Data is upserted into PostgreSQL
  • Database hosted on Railway

Read Path

  • React + Vite frontend
  • Queries only the PostgreSQL database

Benefits

This read/write separation ensures:

  • Even with 100 concurrent users, the backend only queries the database
  • The server IP never interacts with LinkedIn directly
  • Avoids rate limits
  • Provides high stability and performance

2. The 3-Layer Visa Verification Pipeline

Visa classification accuracy was the biggest challenge.

Scanning full job descriptions for the word "sponsor" produced massive false positives.

We designed a three-layer validation architecture.

Layer 1: Sentence-Level NLP

The system:

  • Splits job descriptions into individual sentences
  • Uses regex patterns
  • Detects negation terms such as: cannot sponsor/does not offer sponsorship

This prevents cross-sentence mismatches.

Layer 2: GOV.UK Cross-Validation

The system performs token-set fuzzy matching against the official:

Home Office Licensed Sponsors database

  • 90,000+ companies
  • 85% similarity threshold

This verifies whether the company is officially licensed to sponsor visas.

Layer 3: Composite Verdict

Both sources are synthesized into a 5-tier confidence rating:

Status Meaning
✅ Confirmed Explicitly states sponsorship
🟡 Licensed Company is licensed but not stated
⚠️ Unverified Unclear sponsorship status
❌ No Sponsorship Explicitly states no sponsorship
Not Specified No relevant information

Challenges

Scraping Resilience

Balancing scraping speed and LinkedIn rate limits required careful tuning.

Final configuration:

  • 3-thread concurrency
  • 0.8–2s random delays
  • 6-hour scraping interval

This reduced total scraping time:

  • From 30 minutes (serial)
  • To under 10 minutes

While remaining well under LinkedIn detection thresholds.

What I Learned

Cross-source Data Validation

Combining:

  • GOV.UK official data
  • NLP-parsed job descriptions

created a far more reliable verification system than keyword detection alone.

Importance of Caching Architecture

Separating read and write paths:

  • Solved rate-limiting issues
  • Improved system performance
  • Increased scraper stability

Async System Design

Building a fully asynchronous pipeline helped me understand how Python's event loop interacts with:

  • I/O-bound tasks (scraping, database operations)
  • CPU-bound tasks (TF-IDF computation)

Built With

Share this project:

Updates