We startedwith a business concept to add value to clients for better CX using predictive analytics;as AI expert I had the mandate to handle the project end-to-end

- Successfully developed the AI models using various conventional forecasting algorithms Triple Exponential Smoothing, Seasonal ARIMA as well as theRecurrent Neural Network (RNN) with LSTM for time series forecasting

- Achieved 85pct accuracy in call volume prediction with a variance of ± 10

- Roadmap for further fine tuning to achieve higher accuracy

Approach - Power demand from day to day for a state fluctuates in a relatively steady manner except Sundays and holidays such as festivals, as well as because of rise or fall in temperature. The challenge was to achieve variance of ± 4% (96?curacy). Currently we have delivered a product that achieved a baseline accuracy of 80-85% and identified methodology to correct the remaining predictions.

Next step is to apply regression-based auto correction to deal with abrupt fluctuations during Sundays holidays and zonal temperature fluctuations. All these solutions have been developed by me single handed. The second phase is awaiting development as other priorities came up from multiple clients that was more lucrative to focus on

India Ministry of Tourism has a site that provides for any tourist asking any doubts about travelling in India and tourist places etc. My company had supplied the server and after sales support.

The idea was to enahnce the CX by providing additional automation feature wherein NLP will be used to automatically classify query and provide answer if available or else trasfer to agent

The call center data received from client included various features (i.e. potential factors influencing the chances of campaign sucess - like customer location (tier A city or B ...), customer age group, profession, agent performance track record, etc.

The campaign outcomeswould fall into one of the many possibilities from complete sucess tocustomer opening insurance failure could be complete failure(Not interested to already insured.).

Was able to successfully develop and deliver subject

Theme - MF investors are mostly commoners with little financial expertise. In contrast their investment stakes are big and can mean committing lifetime savings. Therefore, I felt there was a need to develop a tool that’s reliable and simple to help them in taking educated decisions.

Approach - Since NAV is not statistically driven, I used the Equity component of Growth Funds as a parameter that is statistically amenable. The focus was on getting a comparison of predicted performance than getting a precise value. So, I used 7-Day Simple Moving Average (SMA) to represent the trend to predict which fund is likely to perform better.

That was successful and gave upward of 98 pc accuracy on SMA predictionbasis

- Maintance help desk logs received from client had thechallenge of page long verbose paragraphs describing the issue.

- To simplify process first converted it into manageable text -> created Search Utility to pick and return sentences containing the issue.

- These sentences were subjected to Machine Learning to Map the sentences into(verb /noun) phrases to classify à auto allocation (part 2 of the automation process)