Devansh Mody

Developing microservices in Python for integrating social media platforms with products (VN Secure), fetching the data, and storing them in a database.
Building ETL pipelines for social media platforms.
Implementing analysis services for image, audio, video, and text to check for profanities, and track keywords and sentiments.
Developing a profanity checker algorithm using custom regex and later enhancing it by adding a sentence similarity model "sentence-transformers/all-MiniLM-L12-v1".
Designing a queuing system using AWS Lambda, SQS, and Redis for media analysis.
Developing data mining and web scraping solutions for downloading media (image, video, and audio) for social media platforms.
Developing a queuing system for media scrapping using Redis streams, hypercorn, IP and MAC address masking, flask, and DASK framework for handling bulk requests.
Performing image preprocessing to remove background from images.
Implementing a multitouch arbitration model for the marketing team for time series data of influencer campaign.
Building a proof of concept custom model using neural networks based on LSTM and HMM.
Working on a proof of concept on social listening.
Leading the research, design, and development of a content creation platform for influencers using artificial intelligence algorithms and generative AI (LLM).
Identifying personalized content goals and niches from text for influencers.
Developing a custom model with a text pre-processing pipeline specific to social media considering emojis, emoticons, slang, and other important genz trends in text on social media.
Building an image to caption model using nlpconnect/vit-gpt2-image-captioning.
Extracting text and images from all types of unstructured documents using unstructured.io with a custom algorithm for semantic matching, extraction, formatting, and removing PII information.
Working on prompt engineering-based projects with custom prompts for text and image-based content creation services using GPT4-8k.
Implementing realtime conversion of speech to text using Openai-whisper.
Developing Vivian chatbot personal assistant using gpt4-32K, conversational retrieval chain, unstructured.io, and ada embedding model.
Using Chroma db vector database along with Redis and Mongdb.
Building highly scalable architecture using data loaders, streaming, and WebSockets.
Developing video to caption with custom frame selection algorithm using nlpconnect/vit-gpt-2-image-captioning and facebook/deter-resnet-50.
Detecting impersonation in text and images using GPT_4-8k, distilbert-base-nli-mean-tokens, lid.176.bin, mmod_human_face_detector.dat, facebook/detr-resnet-50, and vggface2.
Implementing producer consumer using confluent kafka.
Given any website get the text and images from the website in a proper format.
Building a global trends web scrapper.
Fetching the region and city-wise trending topics.

May 2022 till August 2022:
● Developing microservices in Python for Integrating social media platforms with products (VN Secure), fetching the data, and storing them in a database.
● ETL pipelines for social media platforms.
● Implementation of analysis services for image, audio, video, and text to check for profanities, and track keywords and sentiments.
○ The profanity checker algorithm was developed using custom regex and later enhanced by adding a sentence similarity model “sentence-transformers/all-MiniLM-L12-v1” to the existing algorithm.
● Design of queuing system using AWS lambda, SQS, and Redis for Media Analysis. August 2022 till November 2022:
● ●
● ●
●
Developing data mining and web scraping solutions for downloading media (image, video, and audio) for social media platforms.
A queuing system is developed for media scrapping using Redis streams, hypercorn, IP and MAC address masking, flask, and DASK framework for handling bulk requests.
Image preprocessing to remove background from images.
Multitouch arbitration model implementation for the marketing team for time series data of influencer campaign, proof of concept custom model is built using neural networks based on LSTM and HMM.
Proof of concept on social listening.
November 2022 till date:
● Leading the research, design, and development of a content creation platform for influencers using artificial intelligence algorithms and generative AI (LLM)
● Identifying personalized content goals and niches from text for influencers
○ A custom model with a text pre-processing pipeline specific to social media was developed considering emojis, emoticons, slang, and other important genz trends in text on social media.
○ Models/libraries used are langdetect, yake, language_tool_python, gingerit, inflect, nltk, fasttext, happy transformer, vennify/t5-base-grammar-correction, T5 transformer, fabiochin/t5-base-tag-generation, lid.176.bin
● Image to caption model
○ The model used nlpconnect/vit-gpt2-image-captioning
● Extracting text and images from all types of unstructured documents
○ The model used unstructured.io with a custom algorithm for semantic matching, extraction, formatting, and removing PII information.
● Prompt engineering-based projects with custom prompts for text and image-based content creation services. ○ Model used GPT4-8k
● Realtime conversion of speech to text ○ Model used Openai-whisper
● Vivian chatbot personal assistant
○ ○ ○
Models used gpt4-32K, conversational retrieval chain, uns

Mining data from websites like BSE and NSE, and terminals like Bloomberg and creating pipelines using Kafka to dump data into Cassandra, calculating historical, implied volatility, delta, theta, and gamma, and storing it in the Postgres database.
Building a tool for calculation of SSF OI change which gives a market snapshot by showing some futures contracts that have been changed during market hours.
Developing automation of ADR report for the equity market and automation of options query.
Building a tool for calculation of intra-sector divergence between stocks, finding a correlation between stocks then selecting stock pairs that are above a threshold and classifying them according to 95th and 5th percentile from ratio, and carrying out research analysis with past performance of the same stock pairs and predicting its future performance.
Developing a tool to find dispersion for stocks to help predict a range of possible returns for investments and measure the riskiness of investments.
Developing effective pricing strategies by comparing prices of products sold on websites like Amazon, Flipkart, BigBasket, Reliance Fresh, and Dmart.
Developing software that does daily statistical analysis and generates FI and DI reports for futures, stocks, and options.
Building a tool for automating statistical reports related to finance, FPI clients, equity derivatives, calculation of historical implied volatility and dispersion for stocks.
Performing datamining and web scrapping of daily trends in FI and DI.
Predicting the future of domestic flights for selected destinations by scraping data from Google flights and running predictions on top of it for selected routes to predict future prices.
Calculating the geographic distance between popular retail stores like Dmart and Reliance Fresh in India for statistical analysis.

Developing new features for chatbots and maintaining the existing system.
Developing custom video and text-based chatbot solutions for clients HDFC Bank and loangine.
Designing a data pipeline for moving data between databases.
Performing inference and statistical analysis on data generated by kibana logs for training of chatbot.
Building a data cleaning and text preprocessing pipeline for training chatbots.
Performing ETL and EDA on text, voice, and video data for training and testing chatbot performance.
Building an automation platform for bot performance testing.
Developing a churn model to predict which customers will churn based on their subscriptions, recharge patterns, and usage of services.
Building a text summarization model for Jio care data using seq-2seq transformer model built from scratch.
Performing market basket analysis on the Reliance retail data and building a recommendation system to recommend products that can be brought together.
Developing a POC on fraud detection helping the Jio security team find anomalies or suspicious activity in the data.
Developing API for resetting passwords, changing passwords, bulk user upload, deleting users, creating/deleting admin, and generating/deleting video services.
Developing a service for user profile encryption.
Creating schemas and validation services for schema validation.
Developing an adaptor service so video services of Jio Meet can be integrated and used by the Jio brain bot.
Integrating Jira with Jio brain bot.
Developing a service to fetch data from MongoDB and perform statistical analysis of user information like call history, duration of calls, calls per day/week/month/year, and average call duration.
Validating and testing various services and products of Reliance Jio (Jio chat, Jio Money, Jio Merchant, point of service device, MyJio, Jio.com, Hello Jio, Jio Music, Jio GST, Jio Huddle, Jio Education, Jio Net, Jio Cinema and Jio Cloud).
Serving as team lead for Jio Chat and Jio Net, fixing issues communicating with key stakeholders and managing both applications for quality assurance.