Sep 2024 – Present · 1 yr 9 mos

• 𝗖𝘂𝗿𝗿𝗲𝗻𝘁𝗹𝘆 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗞𝗮𝗶𝘆𝗮 – 𝗔𝗻 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝘁 𝗗𝗮𝘁𝗮 𝗔𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 𝗔𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁
• 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗮 𝗻𝗲𝘅𝘁-𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗮𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 𝘀𝘆𝘀𝘁𝗲𝗺 ("Kaiya"), designed to be a 𝗴𝗮𝗺𝗲 𝗰𝗵𝗮𝗻𝗴𝗲𝗿 𝗶𝗻 𝘁𝗵𝗲 𝗱𝗮𝘁𝗮 𝗮𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 𝘀𝗽𝗮𝗰𝗲. Kaiya empowers users to ask 𝘃𝗲𝗿𝗯𝗼𝘀𝗲, 𝗰𝗼𝗺𝗽𝗹𝗲𝘅, 𝗮𝗻𝗱 𝗺𝘂𝗹𝘁𝗶-𝗶𝗻𝘁𝗲𝗻𝘁 𝗾𝘂𝗲𝗿𝗶𝗲𝘀, intelligently analyzing them to deliver accurate, contextual, and actionable insights.
• 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗱 𝘀𝘂𝗽𝗽𝗼𝗿𝘁 𝗳𝗼𝗿 𝗰𝗼𝗻𝘁𝗲𝘅𝘁𝘂𝗮𝗹 𝗳𝗼𝗹𝗹𝗼𝘄-𝘂𝗽 𝗾𝘂𝗲𝗿𝗶𝗲𝘀, enabling the assistant to retain context from previously asked questions. This allows users to explore data in a natural, conversational flow, breaking down complex analyses into smaller steps while maintaining continuity.
• 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗲𝗱 𝗦𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗥𝗼𝗹𝗲 𝗟𝗮𝗯𝗲𝗹𝗶𝗻𝗴 (𝗦𝗥𝗟) and built multiple 𝗽𝗼𝘀𝘁-𝗽𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴 𝗻𝗼𝗱𝗲𝘀 to refine query interpretation. This ensures better semantic understanding of user intent, leading to more precise query-to-insight transformations.
• 𝗘𝘅𝗽𝗹𝗼𝗿𝗶𝗻𝗴 𝗮𝗻𝗱 𝗶𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗶𝗻𝗴 𝗮𝗱𝘃𝗮𝗻𝗰𝗲𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 𝗮𝗻𝗱 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝘁𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲𝘀 by consuming cutting-edge frameworks and libraries such as 𝗟𝗮𝗻𝗴𝗚𝗿𝗮𝗽𝗵, 𝗖𝗼𝗹𝗕𝗘𝗥𝗧, and vector databases including 𝗖𝗵𝗿𝗼𝗺𝗮𝗗𝗕 𝗮𝗻𝗱 𝗤𝗱𝗿𝗮𝗻𝘁, combined with 𝗘𝗹𝗮𝘀𝘁𝗶𝗰𝘀𝗲𝗮𝗿𝗰𝗵 for hybrid retrieval.
• 𝗖𝘂𝗿𝗿𝗲𝗻𝘁𝗹𝘆 𝗳𝗼𝗰𝘂𝘀𝗲𝗱 𝗼𝗻 𝗲𝘅𝗽𝗮𝗻𝗱𝗶𝗻𝗴 𝗳𝘂𝗻𝗰𝘁𝗶𝗼𝗻𝗮𝗹𝗶𝘁𝗶𝗲𝘀 𝗮𝗻𝗱 𝗶𝗺𝗽𝗿𝗼𝘃𝗶𝗻𝗴 𝗮𝗰𝗰𝘂𝗿𝗮𝗰𝘆, experimenting with advanced query parsing, ranking algorithms, and hybrid retrieval pipelines to push the boundaries of conversational analytics.

Sep 2023 – Aug 2024 · 11 mos

• 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗲𝗱 𝗟𝘂𝗰𝗲𝗻𝗲-𝗯𝗮𝘀𝗲𝗱 𝗲𝗻𝘁𝗶𝘁𝘆 𝗲𝘅𝘁𝗿𝗮𝗰𝘁𝗶𝗼𝗻 𝘄𝗶𝘁𝗵 𝗮 𝘀𝗰𝗼𝗿𝗶𝗻𝗴 𝗮𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺, replacing the earlier tokenization approach. This significantly improved the precision of entity recognition, boosting accuracy to 𝟗𝟓%+ and enabling more reliable Natural Language Search outcomes.
• 𝗘𝗻𝗵𝗮𝗻𝗰𝗲𝗱 𝗳𝗶𝗹𝘁𝗲𝗿 𝘃𝗮𝗹𝘂𝗲 𝗲𝘅𝘁𝗿𝗮𝗰𝘁𝗶𝗼𝗻 by optimizing the search algorithm through 𝗵𝘆𝗯𝗿𝗶𝗱 𝘀𝗲𝗮𝗿𝗰𝗵 𝗼𝗻 𝗘𝗹𝗮𝘀𝘁𝗶𝗰𝘀𝗲𝗮𝗿𝗰𝗵, leading to faster and more accurate retrieval of relevant filters across diverse datasets.
• 𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗲𝗱 𝗲𝘅𝘁𝗲𝗻𝗱𝗲𝗱 𝘁𝗶𝗺𝗲 𝗳𝗶𝗹𝘁𝗲𝗿 𝘀𝘂𝗽𝗽𝗼𝗿𝘁, enriching the product’s ability to handle a wider range of time-based queries. This feature empowered users with greater flexibility in performing temporal analysis.
• 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗱 𝗳𝗼𝗹𝗹𝗼𝘄-𝘂𝗽 𝗾𝘂𝗲𝗿𝘆 𝘀𝘂𝗽𝗽𝗼𝗿𝘁 𝗶𝗻 𝘁𝗵𝗲 𝗧𝗲𝗹𝗹𝗶𝘂𝘀 𝗰𝗵𝗮𝘁𝗯𝗼𝘁, enabling customers to break down complex queries into smaller, sequential interactions. This enhancement delivered 𝗺𝗼𝗿𝗲 𝘃𝗶𝘀𝘂𝗮𝗹𝗶𝘇𝗲𝗱 𝗶𝗻𝘀𝗶𝗴𝗵𝘁𝘀 and improved the overall conversational experience.
• 𝗜𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗲𝗱 𝗮 𝘀𝗲𝗹𝗳-𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗳𝗲𝗮𝘁𝘂𝗿𝗲 𝗳𝗼𝗿 𝗡𝗟𝗣 𝘀𝗲𝗮𝗿𝗰𝗵, allowing users to add custom learning for any word or phrase. This made the search engine progressively smarter and adaptable to business-specific terminology. Additionally, designed and delivered all 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝗱 𝗔𝗣𝗜𝘀 to power this functionality.

Apr 2023 – Aug 2023 · 4 mos

• 𝐃𝐞𝐬𝐢𝐠𝐧𝐞𝐝 𝐚𝐧𝐝 𝐢𝐦𝐩𝐥𝐞𝐦𝐞𝐧𝐭𝐞𝐝 𝐞𝐱𝐭𝐫𝐚𝐜𝐭𝐢𝐨𝐧 𝐚𝐧𝐝 𝐦𝐞𝐫𝐠𝐞𝐫 𝐥𝐚𝐲𝐞𝐫𝐬 𝐟𝐨𝐫 𝐍𝐚𝐭𝐮𝐫𝐚𝐥 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗦𝗲𝗮𝗿𝗰𝗵, successfully removing the dependency on the RASA service. This not only simplified the architecture but also improved the overall accuracy of the system to 95%+, significantly enhancing query interpretation and user satisfaction.
• 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐞𝐝 𝐚 𝐝𝐲𝐧𝐚𝐦𝐢𝐜 𝐛𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐯𝐢𝐞𝐰 𝐦𝐨𝐝𝐞𝐥 𝐬𝐞𝐥𝐞𝐜𝐭𝐢𝐨𝐧 𝐟𝐞𝐚𝐭𝐮𝐫𝐞, enabling models to be chosen automatically based on user-defined rules. This made the system more flexible and adaptable to diverse business use cases, reducing manual intervention and improving efficiency for end users.
• 𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐞𝐝 𝐦𝐮𝐥𝐭𝐢-𝐥𝐚𝐲𝐞𝐫 𝐜𝐚𝐜𝐡𝐢𝐧𝐠 𝐦𝐞𝐜𝐡𝐚𝐧𝐢𝐬𝐦𝐬 and implemented an 𝐑𝐏𝐂-𝐛𝐚𝐬𝐞𝐝 𝐜𝐚𝐜𝐡𝐞 𝐢𝐧𝐯𝐚𝐥𝐢𝐝𝐚𝐭𝐢𝐨𝐧 𝐟𝐥𝐨𝐰, which optimized data retrieval and improved system performance by approximately 𝟑𝟎% . This upgrade reduced latency and delivered a smoother experience for high-volume queries.
• 𝗘𝗻𝗵𝗮𝗻𝗰𝗲𝗱 𝘁𝗵𝗲 𝗲𝗻𝘁𝗶𝘁𝘆 𝗲𝘅𝘁𝗿𝗮𝗰𝘁𝗶𝗼𝗻 𝗽𝗿𝗼𝗰𝗲𝘀𝘀 by building advanced 𝗼𝘃𝗲𝗿𝗹𝗮𝗽𝗽𝗶𝗻𝗴 𝗮𝗻𝗱 𝘀𝗰𝗼𝗿𝗶𝗻𝗴 𝗹𝗼𝗴𝗶𝗰. This refinement allowed the system to extract entities with higher precision, especially in complex queries, leading to better contextual understanding and more accurate Natural Language Search results.