Aug 2023 – Apr 2025 · 1 yr 8 mos · Bengaluru, Karnataka, India · On-site

• At Obviously.ai, I contributed to building AI-driven solutions that extended the company’s product capabilities and generated measurable business impact.
• Collaborated with the VP of Engineering to implement LLM Ops infrastructure, launching generative AI features that expanded offerings beyond predictive analytics and created new revenue streams.
• Developed a Text-to-SQL model using LLMs fine-tuned on database schemas, which reduced query times by 43% and enabled natural language interactions with databases.
• Engineered a clustering model for mail batching, increasing 5-digit saturation percentages and significantly reducing postage costs.
• Increased customer engagement by 34% for health and fitness clients through chatbot deployments using LangChain and OpenAI, improving retention.
• Built a virtual staging model with Stable Diffusion that led to a 17% boost in conversion rates for real estate listings.
• This role sharpened my expertise in LLM applications, MLOps, and production-ready machine learning solutions across multiple industries.

Nov 2022 – Jul 2023 · 8 mos

• 1. Implemented a rule-based decision framework in production that automatically declines loan applications if customers fails in any of those rules, resulting in a 35% reduction in fraudulent loan approvals.
• 2. Designed and implemented credit fraud detection models using XGBoost algorithm in Python, resulting in a notable 20% decrease in fraud rate, saving the company $150k monthly.
• a. Utilized Ray parallel computing and on demand spot instances from Jenkins to speed up the
• training process and reduced the training cost by 70%.
• b. Automated the training process for the fraud models, enabling the team to experiment more
• quickly and catch emerging patterns of fraud.
• 3. Collaborated with cross-functional teams to define key performance indicators (KPIs) and developed weekly reports for monitoring business performance.
• 4. Implemented behavioral models by leveraging historical customer performance data to accurately predict future returns, resulting in a 15% improvement in return on investment (ROI) for targeted marketing campaigns.
• 5. Improved the accuracy of risk assessment by dividing the customers into 10 different segments using classical machine learning techniques. This allowed the company to better identify potential credit risks.

Sep 2021 – Nov 2022 · 1 yr 2 mos

• 1. Implemented risk models to divide the customers into 10 different segments.
• 2. Developed a fraud model pipeline to detect emerging patterns of fraud.
• 3. Incorporated behavioural modeling to inspect customers who were performing poorly.
• 4. Saved 5 hours per week by automating redundant work on Python.

Jan 2021 – Aug 2021 · 7 mos · India · Remote

• Reduced manual inspection time by 50% by constructing a computer vision model using ResNet architecture to detect defective solar panels.
• Deployed machine learning models in production using cloud computing platforms such as AWS Lambda.
• Devised a data pipeline that pulled data from different sources, including the SQL database, resulting in meaningful reports crucial for the weekly macro meetings.