Jun 2023 – Present · 3 yrs

• Working as an AI Consultant to help companies transition into AI-first organizations by guiding them on AI transformation strategies, product architecture, and the adoption of LLM-driven development using tools like LangChain, LlamaIndex, Pinecone, and Vector DBs.
• Working on applying LLMs (GPT-4, BioGPT & SciBERT) to research for interpreting biological datasets
• Fine-tuned domain-specific models and used HuggingFace Transformers, FAISS, and sentence-transformers for embedding for the domains
• Created a RAG framework to integrate various company-internal documents using ChromaDB to enable communicating with those documents in a cost efficient manner
• Experimented with multi-modal LLM architectures and how to normalise the text/image/sound embeddings via CLIP models
• Collaborated with the Supreme Court of India Hacktahon to design an LLM-powered legal engine for document classification, issue tagging, defect identification, and automated summarization
• Built a multilingual legal chatbot leveraging RAG to support the case discussions

Oct 2022 – May 2023 · 7 mos

• Led a cross-functional team of data scientists, engineers, and analysts in developing and deploying state-of-the-art machine learning and AI solutions that directly impact business performance and customer experience. The key focus is on driving innovation and ensuring the seamless integration of advanced technologies into Carousell's core platform.
• One of my primary initiatives is the development of a next-generation search and recommendation engine, built using deep learning algorithms and neural network architectures to provide highly personalized search results and product recommendations, optimizing user engagement and conversion rates at scale.
• Additionally, I am overseeing the exploration of Generative AI models - LLMs in customer support automation and product description generation to streamlined operations.
• As Carousell transitions from a third-party marketplace to a first-party business model, I am leading the design and deployment of a robust price optimization engine. This system integrates predictive analytics, real-time data processing, and dynamic pricing algorithms to enhance profitability and improve pricing strategies based on evolving market conditions.
• Through my leadership, I ensure our data science and business intelligence efforts are aligned with broader C-level organizational goals, fostering a culture of innovation, data-driven decision-making

Mar 2022 – Sep 2022 · 6 mos

• Developed the company’s first comprehensive 2-year data strategy. This initiative brought together the Data Engineering, Business Intelligence, Product Analytics, and Business Analytics teams, providing a cohesive roadmap to improve data governance, enable scalability, and align our data efforts with Carousell’s overall business goals.
• Also initiated the creation of the Data Products discipline, a dedicated team that develops scalable data solutions to improve internal efficiency and enhance user experiences. By establishing this function and appointing a Data Products Manager, we’ve aimed to improve our kkey data products on funnel analysis, experimentation engine.
• Worked with the analytics team for improving personalized recommendations, improved buyer-seller matching metrics, and started dynamic pricing models with the other Carousell's group initiatives — Contributing to Carousell’s increased revenue growth
• Worked with the teams towards integration of AI and machine learning technologies into Carousell’s core platform operations for eg in fraud, optimized pricing, and customer segmentation to provide more relevant and personalized experiences.
• Additionally, I spearheaded the adoption of self-service BI tools and dashboards across the organization, democratizing access to data insights. Worked with several vendors to draft our data cataloging and data governance strategy. This initiative has empowered cross-functional teams to access real-time analytics, accelerating decision-making and improving agility in product iterations. By promoting data literacy programs, I ensured that Carousell remains a data-first company that leverages insights for innovation and long-term success.

Apr 2020 – Feb 2022 · 1 yr 10 mos

• Leading the Product Analytics and Strategy for OLX Europe.
• Building the Product strategy for OLX Europe with a big portfolio of Goods, Jobs, and Services serving millions of users in Europe. Leading the teams in understanding the opportunity space using Customer Journeys, Macro insights, Competitive landscape , drafting the vision, devising various methods of sizing via the top-down and bottom-up method, feasibility and execution strategy, Go-To-Market strategy across 5 countries leading all the way to Roadmap and Goals for the whole OLX.
• Drafting new ways of working within the Analytics team to foster better ways of collaboration, working on cross-functional projects. The projects touch various customer journeys such Transactions, Buyers, Sellers, Growth and help understand the right monetization models, conversion analytics to optimize the product, retention and cohorts, A/B testing, rapid prototyping etc
• Critical cross-functional projects involve building deeper data science models to explore some perplexing phenomenon, building deeper econometrics models for better monetization, building better data models to facilitate quick and consistent analytics on top of it
• Do ping me if interested in any of the roles

Apr 2019 – Mar 2020 · 11 mos

• Leading the Product Analytics teams for Europe and Global teams - which span across all geographies such as Trust and Safety and Monetization teams
• Working with the team and with other functions on building a robust data strategy for the data and analytics functions. Building up a team and drafting career plans, charter, roadmap, and vision for the analytics teams. Worked on the interview process and hiring strategy for the team
• Worked on building the analytics strategy that went behind the rebranding of the platform across multiple channels. Helping the team build forecast models to accurately predict the impact of rebranding and building a prompt feedback loop into our Engineering and Product teams
• Making the budget robust from the data viewpoint

Sep 2018 – Apr 2019 · 7 mos

• Heading the 30+ strong Product Analytics team spanning across multiple locations in Europe (Germany, Poland, Lisbon) and India
• Working with the team on various arms of the product development covering topics for the whole user journey - Seller Experience, Buyer Experience, Trust and Safety, Monetization, Search and Discovery
• Working on building a world-class product by enabling the Product teams to experiment with agility, gauge product success, advanced data exploration and modeling techniques to answer practical business problems and to drive strategy at an executive level.
• Experience of working in strategic projects defining the goals, vision, metrics, frameworks for the whole product team. Additionally integrating them into the budget for the company as well

Nov 2016 – Jul 2018 · 1 yr 8 mos · Singapore · On-site

• Leading the data science efforts for the Strategy and Planning Team
• Building the proactive and real-time technology to resolve complaints even before a user writes
• Building forecasting models for various aspects of the business using even driven time series modelling, via ensemble of ARIMA, ETS and other models
• Developing a platform to monitor tweets at a real time level with sentiment analysis baked into it
• Developing a platform to analyze the global media channels and their perceptions about Uber using Google BigCloud, Google Big Query, Python and various NLP Techniques
• RNNs, Bag of words based analysis of various texts to compute the key phrases used by the customers by constructing various N-grams and word associations( Natural Language Processing )

Nov 2013 – Nov 2016 · 3 yrs · Delhi, New York, United States · On-site

• Quantitative Modelling
• Building models for evaluating the margin requirements for the hedge funds in the execution and clearing business
• Modelling the quantitative risk of equity and credit derivatives, convertible bonds and synthetic products
• Risk Management for Hedge Funds
• Slicing and dicing of hedge funds portfolios to predict the risk posed by the market fluctuations in case of extreme events
• Building risk models to allow hedge fund managers to execute a wide range of investment strategies, which are immune to the market volatilities

Nov 2011 – Oct 2013 · 1 yr 11 mos · Bangalore · On-site

• Worked on customer segmentation, revenue forecasting, and inventory optimization, with a focus on delivering actionable insights for Product teams
• Key Projects & Responsibilities:
• 1. Customer Segmentation & Retention Models:
• I developed and deployed unsupervised learning models like K-means clustering and Hierarchical Clustering to segment Myntra’s customer base by behavior, demographics, and purchasing patterns. These insights led to more targeted retention campaigns, increasing customer retention by 15%. Additionally, I built models to predict customer lifetime value (CLV) and churn, enabling focused strategies for high-value customers and reducing churn rates.
• 2. Sales and Revenue Forecasting:
• I designed time series forecasting models using ARIMA, SARIMA, and Facebook Prophet to analyze revenue trends and predict sales fluctuations. By incorporating seasonal and external factors, my models improved forecast accuracy by 10%, informing better inventory management, pricing, and marketing strategies.
• 3. Inventory Optimization:
• I developed a catalog model using multivariate regression and XGBoost to understand the impact of inventory on sales. By identifying correlations between stock levels and sales performance, my model optimized procurement and reduced overstock by 20%, enhancing inventory turnover and overall profitability.
• 4. A/B Testing & Experimentation:
• I conducted A/B tests to assess the impact of pricing, marketing, and product initiatives. Using Bayesian optimization and hypothesis testing, my insights led to a 12% improvement in conversion rates.

Apr 2011 – Oct 2011 · 6 mos · Seattle, Washington, United States, India

• Worked on optimizing pricing systems that support millions of products across multiple geographies, building robust algorithms and automating key processes in the pricing logic - approvals, overrides, competitive monitoring etc
• Developed and maintained pricing algorithms to handle real-time cost fluctuations for diverse product categories, ensuring optimal pricing across various geographies. Used statistical models to account for currency exchange rates, local taxes, supply-demand dynamics, and competitor pricing.
• Implemented flexible pricing models to adapt to regional variations and product-specific
• strategies, enabling consistent and data-driven pricing decisions at scale.
• Collaborated with data science teams to integrate machine learning models that forecast pricing trends and recommend optimal pricing windows.
• Led the development of an automated price alert system using a Dojo-based framework to monitor price changes in real-time and notify stakeholders about significant fluctuations. Streamlined the process of flagging discrepancies between cost and retail prices, reducing manual intervention and improving pricing accuracy across product lines.
• Designed dashboards and monitoring tools using Dojo and React to visualize pricing trends and ensure seamless integration of alert mechanisms with existing systems.
• Utilized Java, Spring Boot, Node.js, React, Dojo to build and deploy scalable services for Amazon’s global pricing platform.

May 2010 – Apr 2011 · 11 mos · On-site

• Worked on building and scaling digital platforms to meet the complex financial needs of High Net-Worth Individuals (HNIs). Created data-driven solutions to empower portfolio managers/risk managers in Switzerland to efficiently oversee and enhance client portfolios
• Developed and enhanced of a web-based platform for portfolio management, tailored to the unique requirements of HNIs. By collaborating with PMs, RMs, I ensured that the platform is equipped to handle the intricacies of managing complex portfolios, offering real-time insights and analytics for better decision-making.
• Also created a scalable report generation tool utilizing XSLT, XML, and VBA frameworks, designed to handle data-intensive reports exceeding 100MB that can't be handled via Excel. This tool improved the efficiency of our reporting process, cutting down the time taken to generate detailed client portfolio reports while ensuring data accuracy and reducing manual intervention.
• I also supported strategic initiatives aimed at client retention and acquisition via various analyses and reports

Mar 2010 – Aug 2010 · 5 mos

• Worked on estimating population growth rates in the U.S. and predicting the likelihood of corporate bankruptcy.
• Modeled population growth using linear regression models and non-linear methods. Used MLE to create a linear model for estimating growth rates based on historical census data and key demographic data. This model provided estimation of the population's trajectory over time
• Also used non-linear techniques from my Master's Thesis such as polynomial models and kernel density estimation (KDE). By applying non-linear kernels to the data, I was able to enhance the model's accuracy in capturing more complex growth patterns that the linear models might miss, especially in regions experiencing unusual demographic shifts. This approach allowed for more accurate predictions of population trends in both the short and long term, providing a deeper understanding of demographic changes and enabling better forecasting for policy-making and resource allocation.
• In the second project, I developed predictive models to assess the likelihood of impending bankruptcy for firms. Using balance sheets, cash flow statements, and profitability ratios, I tried both linear regression and logistic regression models to identify early warning signs of financial distress. The model incorporated key variables like debt-to-equity ratios, operating margins, and liquidity indicators to flag companies at high risk of default.
• For a more robust prediction, I also implemented machine learning techniques such as Random Forests and Support Vector Machines (SVMs) to improve prediction accuracy. These models performed better in terms of identifying patterns that led to financial instability. Validated these models using cross-validation techniques
• Through this internship, I gained hands-on experience with regression analysis, maximum likelihood estimation, and non-linear kernel methods, while applying these techniques to practical challenges in demographic analysis and corporate finance.

May 2009 – Jul 2009 · 2 mos · On-site

• Enhanced the syslog-ng logging infrastructure for VoIP-based products, crucial for monitoring and troubleshooting real-time communications systems. My key responsibilities involved optimizing the logging system to improve data collection, parsing, and storage for more efficient and scalable performance.
• Refined the syslog-ng configuration to improve filtering, forwarding, and storage of log data. By adjusting logging levels (error, warning, info), I enhanced the system’s ability to capture critical events while reducing unnecessary log traffic. Signal to Noise ratio had to be maximised
• Implemented custom log parsing rules using regular expressions and pattern databases to better analyze VoIP logs, enabling faster identification of issues such as SIP call failures, packet loss, and jitter.
• Studied SIP (Session Initiation Protocol) and RTP (Real-Time Transport Protocol), that basically ensure the logging system captured key data like call setup, termination, and quality metrics. This provided deeper insights into VoIP performance for faster diagnostics and patterns to look out for in the logs.

May 2008 – Jul 2008 · 2 mos · On-site

• contributed to the enhancement of validation tools for Intel's smallest and most power-efficient processor, Atom. During my 3-month internship, I worked on optimizing validation workflows and improving the efficiency of validation processes.
• The key area of my internship was to analyze the impact of branch instructions on the overall validation time of the Atom processor. By identifying patterns in the validation cycle, I helped reduce validation time, which contributed to faster testing and debugging processes. Collaborated with senior engineers to optimize validation algorithms and ensured that branch instructions were effectively integrated into the validation toolchain, ultimately improving the performance of the validation tool.
• Additionally, I supported the debugging of validation issues and gained hands-on experience with Intel’s proprietary tools, enhancing my understanding of processor architecture and the complexities of the validation process.