Naman Mohnot

Naman Mohnot

Machine Learning Intern

Followers of Naman Mohnot434 followers
location of Naman MohnotGurugram, Haryana, India

Connect with Naman Mohnot to Send Message

Connect

Connect with Naman Mohnot to Send Message

Connect
  • Timeline

  • About me

    Senior Product Engineer, Applied AI, Sprinklr | CSE, IIT Madras

  • Education

    • Delhi Public School, Jodhpur

      2015 - 2016
      Matriculation CBSE 10 CGPA
    • Indian Institute of Technology, Madras

      2018 - 2022
      Bachelor's degree Computer Science
    • Jayshree Periwal High School, Jaipur

      2016 - 2018
      Intermediate +2 Science and Maths 94.4%
  • Experience

    • Ayu Devices Pvt Ltd

      Dec 2019 - Jan 2020
      Machine Learning Intern

      Successfully classified heart sounds as normal/abnormal using an LSTM model in TensorFlow, leveraging pre-processing techniques such as Hidden Markov models, and enhancing accuracy with noise reduction using signal processing algorithms

    • Sprinklr

      May 2021 - Jul 2021
      Summer Intern
    • Sprinklr

      Jul 2022 - now

      ๐—ฆ๐—บ๐—ฎ๐—ฟ๐˜ ๐—–๐—ผ๐—บ๐—ฝ๐—ผ๐˜€๐—ฒ:โ€ข Led the development of a Gen-AI enhanced "typing aid" feature to provide precise suggestions for customer support agents. โ€ข Innovatively created a high-performing scoring mechanism using causal LMs and brand data, enhancing precision twofold. Utilized optimization strategies such as mixed precision, kernel fusion, ONNX, ONNX Runtime, and the MQTT protocol to reduce system latency by more than 50%. Scaled this system from 20 to 500+ clients within a week.๐—ฆ๐—บ๐—ฎ๐—ฟ๐˜ ๐—ฅ๐—ฒ๐—ฝ๐—น๐˜†:โ€ข Created an SOP-grounded next response generation feature, assisting customer case agents.โ€ข Utilized a "patent-pending" method to derive agent handbooks and knowledge bases from conversation data via clustering and LLMs. Employed advanced optimization tools such as MLC and TensorRT, leveraging quantization, precision optimization, and hardware acceleration. โ€ข These methods enabled in-house models to achieve 95% accuracy at 10x lower cost compared to GPT-4 and generated a new revenue stream through integration into Sprinklr's AI+ suite.๐—–๐—ต๐—ฎ๐˜-๐—๐—ฎ๐—ถ๐—น๐—•๐—ฟ๐—ฒ๐—ฎ๐—ธ๐—ฒ๐—ฟ: โ€ข Developed a novel โ€œJailbreaking Chat Simulationโ€ technique to train, test, and evaluate Gen-AI bots. Leveraged powerful LLMs to simulate conversations aimed at breaking guardrails, generating a "jail-break proof" dataset. โ€ข This significantly enhanced model adherence to guardrails and knowledge bases. Boosted model accuracy from 65% to 95%, facilitating the full transition of voice bot volume to the new model. Automated testing reduced time from a full day to 1-2 hours, greatly improving efficiency and expediting partner onboarding. Show less

      • Senior Product Engineer

        Jun 2024 - now
      • Product Engineer

        Jul 2022 - May 2024
  • Licenses & Certifications