Data Science Salon Seattle is a virtual event where AI researchers, data scientists, and e‑commerce professionals can exchange experiences and ideas without the need to travel or relocate. In doing so, the event follows the DSS mission to make the latest technologies more accessible and inclusive for every talented individual, regardless of background.
Traditionally, the Seattle event focuses on AI applications in e‑commerce. Invited experts and practitioners share their insights so participants can broaden their knowledge and apply it in the businesses they operate.
The opening talk was delivered by Medha Gupta, Senior Data Engineer at Beyond, Inc. She explored how a company can design scalable data pipelines, use AI models for demand forecasting, and ensure low latency on a global e‑commerce platform.
“AI has become a game‑changer for pricing strategies,” Medha said. “We feed pricing engines with internal data such as past transactions and customer behavior, and we enrich it with external data,” she added.
Jitender Jain, Sr. Manager / Staff Software Engineer at Walmart, discussed turning retail challenges—like stockouts and excess inventory—into business advantages through smart AI technologies.
“Stockouts cost retailers over $1 trillion globally, and excess inventory locks working capital,” he said. Fragmented systems and siloed insights hamper retailers’ ability to forecast required inventory, he added.
Jinal Mehta, Data Engineer II at Amazon, detailed the supply‑chain data architecture built for the world’s largest retailer. The key challenge—and opportunity—was designing a robust, reliable data pipeline.
Arthur Keen, Senior Solution Architect at ArangoDB, focused on battling counterfeit goods in marketplaces. Unethical sellers pose multiple risks—lawsuits, disappointed customers, and brand damage. AI‑driven monitoring is one of the few effective ways to manage the problem.
“Counterfeit goods on marketplaces are a bigger problem than I ever imagined,” he said. “They cost more than $300 billion a year, with 50 percent of brands experiencing losses.”
Panelists: Jacob Marquez (ConnectTo), Nandita Krishnan (Adobe), Tejaswini Sirlapu (Intel), and Rossella Blatt Vital (Sprout Social). The experts discussed how AI is reshaping modern retail and e‑commerce, and how to measure success for today’s AI deployments.
Can LLMs Forget What They Learned? Risks from Memorization and Mitigation Strategies
Sriram Selvam, Senior Software Engineer at Microsoft, examined how large language models can both learn and unlearn sensitive or copyrighted information, allowing companies to mitigate risk.
Amit Dhanda, Senior Scientist at Amazon, showed how machine learning transforms finance—a data‑rich sector full of numerical and textual information. He demonstrated AI use cases such as automated document handling.
“The data landscape in finance is pretty rich,” he noted, “but it’s fragmented, with some parts in Excel and others in PDF financial reports.”
After a short break, Srinivas Murri, Lead Data Engineer at Meta, discussed using synthetic data to accelerate model development and lower costs.
“Data is expensive,” Srinivas said. “Cheaper, synthetic data can be used for testing, validating, and training models.”
Apurva Kumar, Principal Software Engineer at Walmart Global Tech, demonstrated how AI agents operate autonomously within Walmart’s tech ecosystem, reducing manual errors.
“The AI infrastructure is the backbone of modern applications like demand forecasting or personalization, yet it’s surprisingly manual behind the scenes. AI agents help avoid human mistakes,” he explained.
In the closing session, Bhargav Jairam Shetgoankar, Senior Data Scientist at Target, revisited classical time‑series methods for today’s AI‑driven retail.
“Retailers gain an edge by forecasting accurately,” Bhargav said, adding that the catastrophic effects of missed predictions were evident during the COVID‑19 pandemic.
The event offered a perfect opportunity to exchange knowledge and network with fellow data‑science professionals—without leaving home or office. Recordings will be available on demand for two weeks after the conference.