ROHIT POTE

• I have worked with Darwinbox in various capacities including as a freelancer, intern, and as a full-time junior data scientist. Throughout my tenure, I successfully completed numerous projects by thoroughly understanding the business requirements, applying machine learning and AI insights, and meticulously attending to other critical details which propelled my projects into production. Additionally, my contributions earned me the highest appraisal and significant recognition within my team.

Projects:

• Text Classification: Trained a large language model (LLM) on labeled employee engagement comments data with over 31 predefined labels. The trained model was deployed in production to predict labels for new comments.
• Analytics AI Chatbot: Developed an AI-powered chatbot using LangChain for context retention, enabling seamless conversations. The chatbot responds to user queries in both textual and visual formats, dynamically generating relevant data visualizations for enhanced insights.
• Sentiment Analysis: Implemented a sentiment analysis model using the RoBERTa model pretrained on Twitter comments. This model, deployed in production, predicts the probability of the sentiment of comments.
• Clustering Analysis: Utilized KMeans clustering to group employee engagement data into different clusters. Employed OpenAI to generate insights about these clusters, helping to understand the concerns of different employee groups.
• JD Generator: Developed a job description (JD) generator using OpenAI, taking minimal inputs from users to avoid biases and hallucinations through well-engineered prompts. Skills generated by OpenAI were matched to company-standard skills from a skill ontology database using embeddings and cosine similarity. This solution is deployed in production.
  
• ttrition Prediction: Built two types of attrition prediction models: a global model with common variables across different tenants, and a tenant-specific model with higher accuracy tailored to particular tenant data.
• Search and Matches: Extracted named entities (NER) from JDs and resumes, using cosine similarity of embeddings to find top-matching resumes for JDs. Initially used a BERT model trained on resume and JD NER, and later experimented with the LLaMA 2 model. OpenAI was used to provide insights on the alignment of NER between top-matching resumes and Jds.
• Buddy Recommendation System: Developed a recommendation system to match new joiners with buddies for initial onboarding assistance. Employed a sentence transformer model to generate embeddings and FAISS to find the top 10 matching buddies. This system is deployed in production.

• Started my career in the email and chat support department, where I developed a strong understanding of customer service by effectively handling inquiries from irate customers
• Demonstrated exceptional performance in the chat process- highlighted by my quality score and average handling time - which led to a promotion to the email process within the first six months.