Hire Ashok P. - NLP Engineer

Problem Statement: Working as a NLP engineer in regulatory management system where the business is mostly related to implementing rules which are imposed by countries. Project has three phases: First Phase:Thomson routers sends more than 1000 feeds per day from web , based on summary, we designed the first binary classification model to categorize it as impactful or not. Involved in all phases of pipeline of model development. Preprocessed the data with lamatization and other techniques. Explored different options like BERT,GPT and FAST. Observed the cosine similarity of words on trained model to get an idea. Based on the metrics and evalution we opted Fast and trained with domain specific data and achieved the 76% of accuracy. Analysed the F1Score,recall and precision to make sure model is not baised to any category. Implemented the monotoring to observe drift and behaviour of model in different information. Second Phase:All impactful feeds, needs to get process and system should smart enough to mitigate which line it says or much need full info for rules implementation. Higher management has an impression that, in this phase we need to have some model where it can have ability to read the pdf and generate citaions. As a developer, I asses the aaproach and recommnded AWS Textract and Adone PDF extract to get the citations with structure. After all discussion s got approval for Adobe PDF extract option and developed the program to generate a citations by taking Adobe output json as input. Third Phase:In MVP3 phase we concentrated on category determination of each citation by using NLP Bert technique and feeded the archer system with recommendations. It is multiclass classification for 12 categories. We recommended stake holders to come up with 3 broder categories instead of 12 to get better results and recommendation. Worked on the imbalance data set by using ensemble technique,smorte both are not given good results.

* Problem Statement: Sell My Car feature has to give instant appraisal of car value by analysing condition report which was submitted by customer.Third party(inventory + team) is providing the price by manual intervention and it takes 30 mins to send the price certificate to customer through email.Analyse on data which was available from past 5 years and predict the price, so ROI will increase and gives best satisfaction to customer. * Responsibilities: * o Utilized various techniques like Histogram, bar plot, Pie-Chart, Scatter plot; Box plots to determine the condition of the data. * o Worked on data processing on very large datasets that handle missing values, creating dummy variables and various noise in data. * o Performed data pre-processing tasks like merging, sorting, finding outliers, missing value imputation, data normalization, making it ready for statistical analysis. * o Implemented ridge regression, subset selection methods to choose most statistically significant variables for analysis. * o Used various machine learning algorithms such as Linear Regression, Ridge Regression, Lasso Regression, Elastic net regression, KNN, Decision TreeRegressor, SVM, Bagging Decision Trees, Random Forest, AdaBoost, and XGBoost. * o Build machine learning processes that monitor and provide feedback on how or where to improve predictive models already deployed. * o Work closely with business stakeholders, Financial Analysts, Data Engineers, sound organizational decisions.Data Visualization Specialists and other team members to turn data into critical information and knowledge that can be used to make the prediction successful. * o Created ML Pipelines using Azure Functions and Azure Web. * o Data Collection from different sources and loading into destination table. * o Creating ADF Pipelines and necessary Data flows to transform the data. * o Json parser implementations,derived columns for further analyzation and model input.