Darpan Jain

Pixstory

Lead Machine Learning Engineer

• Heading development of Pixstory's Retrieval Augmented Generation (RAG) based Conversational Search, accurately answering user queries by grounding responses in a verified knowledge base and generating nuanced, de-opinionated answers.
• Mitigating hallucination issues in large language models and addressing misinformation by ensuring reliable information retrieval through rigorous source vetting and verification. The retrieval process is re-ranked based on the level of trustworthiness of the sources.
• Public beta launch upcoming, featuring an immersive and user-friendly interface. Stay tuned for updates!

(May 2023 - Present)

Information Sciences Institue (USC-ISI)

Applied Researcher

• Led development of conflict resolution dialogue agents for DARPA’s Civil Sanctuary Program, as an Applied Researcher with the Natural Language group (CUTE LAB NAME) at USC-ISI.
• Designed and deployed a scalable chatbot for mitigating toxic behavior on social media through non-violent communication strategies, including prompting GPT-3’s advanced transformer-based (the model that drives ChatGPT!) generation capabilities.
• Moderated over 15k multilingual posts with 85% true positive rate in French and German on popular subreddits.
• You can find the published research paper here, which will presented at NAACL 2024.

(January 2022 - July 2023)

Lumin.ai

Software Development Intern

• Contributed in developing the conversational AI for Lumin.ai as a Smart Scheduler & Sales Accelerator to enhance customer interactions and optimize the sales funnel for 5+ product and franchise owners.
• Advocated for efficient documentation using Markdown and managed customer interactions for multiple clients.

(May 2022 - August 2022)

Warner Bros Discovery

Software Development Engineer - Machine Learning

• Boosted user engagement on Discovery Plus (D+) as part of the ”Personalization and Recommendations” team.
• Employed Apache Spark and Airflow to build efficient ETL pipelines in a distributed environment, and executed A/B tests to validate and refine new recommendation features for the D+ platform.
• Led the development of server-side ad-insertion SDKs for Android and Web, driving increased ad revenue for D+ sports content in Europe.

(September 2020 - August 2021)

AdSparx Inc

(now Warner Bros. Discovery)

Machine Learning Engineer

• Spearheaded the research, design, and development of MiDAS, a micro-service-based system serving personalized ads to millions of users on OTT platforms.
• Designed and trained Computer Vision models with TensorFlow, deployed on Kubernetes to enable dynamic server-side ad-insertion for 40+ channels, boosting revenue for U.S. publishers.
• Developed a tool to automate the dataset annotation process, by leveraging SCTE-35 markers in HLS and DASH streams.
• Leveraged multi-step recurrent networks on time series data for ad break forecasting. This resulted in a 30% reduction in resource consumption and network calls to the SSAI backend.
• Member of the senior development team at AdSparx that was acquired by Warner Bros Discovery in September 2020.

(March 2019 - September 2020)

IoTIoT

Machine Learning Engineer

• Led and mentored a team of 30 for catering Machine Learning applications with on-Edge capabilities, for Embedded Linux platforms.
• Developed a crowd flow analysis product using face and voice recognition for event registration, optimized for on-chip GPUs to showcase AI-on-Edge computing capabilities.
• Worked on Pose estimation for detection and tracking of human body pose to create an immersive experience for users for a video gaming platform.

(May 2018 - January 2019)

Defeat COVID-19

Machine Learning Lead

• In light of the COVID-19 pandemic, co-founded DefeatCovid, a not-for-profit organization, which leveraged technology to increase public awareness and aid in curbing the spread of the disease.
• Led a team to build a tool to provide early stage diagnosis of the symptoms and risk assesment for Covid-19, using crowdsourced data and Deep Learning techniques.
• The tool included a Question-Answer Chatbot based on BERT to help navigate users through the website, provide information from government verified sources and address misinformation about Covid-19.
• It also included a real-time dashboard and interactive heatmap interface to identify at-risk localities.

(March 2020 - May 2020)

Prasaurus Sports Analytics

Artificial Intelligence Consultant

• Led a team to design a framework for real-time action recognition and perform statistical analysis, providing deep insights for player performance tracking and improvement, for an academy training national badminton players.
• The models captured Spatio-temporal features across frames based on two-stream networks and ActionVLAD for Action Recognition, and shuttle and player tracking using a custom trained MobileNet, to provide metrics such as 'Dominant areas'.

(June 2019 - September 2019)

InProspect Technologies

Machine Learing and Strategic Consultant

• Managed a team of engineers to design and develop a product to propel undergraduates towards career goals by providing curated career guidance. The platform provided analytics for educational institutions via the 'Enterprise Resource Planning' system.
• Created a personalization system for recommendations to help achieve user-specified goals by aiding in coursework selection and extra-curricular activities. This ensured high success rates in the individuals' field of interest.
• Was part of the strategic planning for incorporating product changes to drive user engagement.

(March 2019 - May 2019)

Ubuntoo LLC

Content Contributor

• A global environmental solutions platform promoting sustainable economic growth.
• Researched and created technical reports on companies and organizations innovating in the use of Machine Learning and Computer Vision.
• The solutions included high accuracy waste sorting systems and autonomous waste collection bots; centered around using technology to tackle environmental issues and promote sustainable development.

(September 2019 - December 2019)

Occipital Tech

(now AgroGrade)

Computer Vision Engineer

• A startup offering products in Agrotech to classify and grade agricultural produce using Computer Vision, resulting in reduced wastage and maximizing profits.
• Worked on the pipeline for preprocessing of real-time image data by Background Segmentation using methods like Gaussian Mixture Models and Watershed Algorithm to tackle camera jitter and dynamic lighting conditions.
• The effort led to an ~20% reduction in preprocessing time, along with enhancing image noise reduction by 14-15%.

(February 2019 - March 2019)