Apr 2024 – Present · 1 yr 11 mos

• Growth:
• Propensity models for Cross Sell, Retargeting different vas services
• GenAI / Search & Personalization -
• Led the in-house intent and entity detection architecture development for AI Powered chatbot " Autonaut",
• Developed RAG system using proprietary knowledge base leveraging vector db and language model embeddings,
• Created Car Service History report consisting of accidental history, service history at service date, odometer, service parts details from unstructured pdfs varied across different OEMs,
• Led the development of Search autosuggestion generation, scoring and ranking module,
• Assured Buyback:
• Led the development of Pricing Algo for predicting car depreciation and forecasting future buyback price basis demand, supply, car cohort risk, customer risk. Optimized via loss provisioning along with car eligibility guidelines basis condition.
• Integrated it for new and used cars partnering with different channel partners across OEMs, financial institution, Insurers.
• Technology stack: GPT, Vector Embeddings, VectorDB, Vector Search, Retrieval-Augmented Generation (RAG), NER, Transformers, BERT, Large Language Models (LLM), Prompt Engineering, LangChain

Apr 2022 – Mar 2024 · 1 yr 11 mos

• Building User Personalization for used car buyer platform. Solving for faster car discovery with personalized recommendations for each user.
• Solving Search Query Resolution with state of the art NLP technique via spell correction+phonetics > Entity Detection > Semantic Embedding generation
• Leveraging LLM/ChatGPT/OpenAI & Transformers for conversational search
• India Cars Sell Online: Enabling customers to sell their cars fastest way by just filling basic car details and answering car condition related questions based on which He is provided a binding offer instantly and can choose to sell car at that price. Worked on from ideation of questionairre to building pricing and flagging model enabling improvement in funnel metrics and higher conversion rate with better CX.
• Technology stack:
• Development Tech Stack: Python, PySpark, MySQL, Bigquery, Snowflake, PubSub
• Visualization: Matplotlib
• Platform: AWS, GCP
• Algorithms: GBDT, Catboost
• Deployment: Flask | meinheld | gunicorn | Bitbucket | Teamcity

Feb 2021 – Mar 2022 · 1 yr 1 mo

• Leading a team of 4 junior data scientists in building robust AI/ML driven solutions for challenging business problems.
• Key Projects :
• India Cars Sell Online: Enabling customers to sell their cars fastest way by just filling basic car details and answering car condition related questions based on which He is provided a binding offer instantly and can choose to sell car at that price. Worked on from ideation of questionairre to building pricing and flagging model enabling improvement in funnel metrics and higher conversion rate with better CX.
• Fraud Detection Engine: Identifying and preventing fraud to prevent losses and reduce cancellation rate by dealer partners. Brought down cancellation rate from ~20% to ~10% and high loss rate from 3% to 1.2% via flagging the car basis car related params, customer data, dealer data and detecting customer-dealer, dealer-dealer collusion in auction process.
• Used Cars Procurement Pricing Engine across AU, UAE & INDIA
• End Buyer Pricing for procured cars: Created B2C pricing model to predict end buyer price of a procured and refurbished cars. This made us more accurate in procurement process and enabled business to make double digit margin.
• Preowned Bikes Procurement Pricing Engine
• Productionizing:
• Microservices for the productionized ML models using Flask|Gunicorn in python, hosted on AWS
• Deploying models via google cloud functions
• Technology stack:
• Development Tech Stack: Python, Mongodb, MySQL, Tensorflow, Keras
• Visualization: Matplotlib
• Platform: AWS
• Algorithms: XGB, LightGBM, ssd_mobilet, ResNet50, LSTM
• Deployment: Flask | meinheld | gunicorn | Bitbucket | Jenkins

Mar 2020 – Feb 2021 · 11 mos

• Built Fraud Detection Engine to prevent high loss cars and reduce cancellation rate,
• RI change model to identify change is two versions and inspection report and detect whether the changes where downgrades in car condition and would that lead to cancellation of active dealer bid on car.
• Developed NumberPlate Masking model where for any given car image model would detect the numberplate and mask it with Cars24 logo. This was previously done by manually placing Cars24 Logo playcard on top of numberplate to keep customer identity hidden and cars24 branding. This led to reduction in inspection time by 10% and saving 4 min per manual exercise.
• Used Cars Procurement Pricing Engine for India and UAE

Oct 2018 – Feb 2020 · 1 yr 4 mos

• Used Cars Procurement Pricing Engine for India
• Online Webquote pricing model for getting overview of car price on different conditions viz. Fair, Good, Very Good, Excellent.
• ML based condition rating engine for 5 different subsection of cars viz. Engine&Transmission, Exterior&Tyres , Electricals&Interior , AirConditioning , Steering&Suspension
• CNN based classification model to predict car color and tyre tread depth
• Audio file processing to detect type of engine noise( Normal, Tappet noise, Timing noise, Bearing noise, whistling noise etc.)
• Productionizing:
• Microservices for the productionized ML models using Flask|Gunicorn in python, hosted on AWS
• Technology stack:
• Development Tech Stack: Python, Mongodb, MySQL, Tensorflow, Keras
• Visualization: Matplotlib
• Platform: AWS
• Algorithms: XGB, LightGBM, ssd_mobilet, ResNet50, LSTM
• Deployment: Flask | meinheld | gunicorn | Bitbucket | Jenkins