A Kivy based application to determine the fitment of an employee from the supply database for particular demand heuristics. A Fitment Score is calculated and based upon which the employee is categorized with the Fitment Segment, namely:

1. Best Fit (85 and Above)
2. Stretched Fit (70 to 85)
3. Best Bet (60 to 70)
4. No Segment (Below 60)

A Score is calculated for each of the demand heuristic compared to the supply's attributes and the Fitment Score is calculated for the weightage given to the heuristic summing to 100%, i.e., the final Fitment Score. The Heuristics considered for the calculation of Fitment Score are namely:

1. Location of the employee(City, Country): The great-circle distance between the demand location and the employee's location and a decrement of weight assigned per 1000km from the demand location.
2. Years of Experience of the employee: A Perfect score (weightage of attribute) for match, while a 10% decrease of the weight assigned for every year short of the demand and a 5% increase of the weight assigned for every year exceeding the demand.
3. Hierarchical Rank of the employee: A perfect score for same rank as demand, while a 10% decrease for every rank lower than the demand and a 5% increase for every rank higher than the demand.
4. Bench Aging of the employee: The number of weeks spent on the "Bench", i.e. number of weeks since not assigned to any project. A negative 10% decrement of assigned weight for every week spent on the bench.
5. Skills of the employee: The specific type of skill requirements of the demand are compared to all the same type of skills of the employee with a word embedding algorithm and the maximum matched skill value is taken along with the weight of the heuristic to calculate the score. Types of skills considered:
   1. Technical Skills of the employee.
   2. Functional Skills of the employee.
   3. Process Skills of the employee.

For the skill matching algorithm we have trained our own skill corpus using the open sourced gensim library, and its Word2Vec Bag-of-Words model. The learnt 100-dimensional embeddings are compared (Demand to that of Supply) using cosine similarity. Score is calculated by mutlipying the normalized cosine similarity with skill levels of the supply, and then choosing the max matched skilled. Download thefolder embedding vectors from Word Embedding Models

Once the Fitment Scores and Segments are determined for each employee in the supply, the recommendation for the demand are ranked in each Segment according to the hierachical structure of the oraganization, i.e., Service Line -> Sub Service Line -> SMU/Business Unit. Implying the employees in the same SMU, SSL and SL are ranked higher while the employees with same SSL and SL but different SMU are ranked lower within each segment.

The dataset acquired from the Hackathon page is extrapolated provided by the host themselves. The key sheets in the dataset are:

1. Demand Sheet
2. Supply Sheet ->
   1. Non-Skills Supply
   2. Skills Supply
3. Skill Tree

1. Pandas
2. Numpy
3. Kivy
4. GenSim

Do make the window of the application full-screen for best experience

conda env create environment.yml
conda activate ey
python ey.py

Do wait a while for the result page to load after clicking the submit button on Weights Page.

To run from github repository

git clone https://github.com/aaditagarwal/Fitment-Predictor.git

Download the trained word embedding from Drive Link
Extract the files from the .zip file and store in "models" and "dataset" sub-directories respectively in the root directory

conda env create environment.yml
conda activate ey
python ey.py

1. Demonstration Video
2. Presentation

1. Aadit Agarwal
   ABV-IIITM, Gwalior
   agarwal.aadit99@gmail.com
2. Aitik Gupta
   ABV-IIITM, Gwalior
   aitikgupta@gmail.com