Sep 2023 – Aug 2025 · 1 yr 11 mos · Gurugram, Haryana, India · Hybrid

• Working on building the in-house GenAI products in speech, NLP & CV domain.
• 1. Cars24 Virtual studio
• Transforming raw car images into professional studio-quality visuals, End to end experimentations,
• training, inference optimisation and efficient deployments.
• 2. speech Processing & Speech2Text:
• Fine-tuning whisper for speech to text translation for Indic Languages outperforming current SOTA model available commercially and open-source.
• Efficiently Deployed Whisper using KFserving in production with inference batcher, capable of
• transcribing 2400 hours of audio data in a day.
• 3. LLM-Driven Call Agent Support System for Enhanced Customer Interaction:
• ◦ Developed an end-to-end pipeline and streamlit dashboard for call analysis, aiding agents by summarizing
• past interactions with customers and facilitating better guidance (solving one agent per customer).
• ◦ Developed car sales pitch tool utilizing Cars24 catalogue, providing key selling points, pros & cons from
• trusted websites using the fingerprints of the car to agent. Leveraged web scraping tools like serper.dev API
• and LLMs for summarization.
• 4. Computer Vision: Cars24 Used Car Inspection Automation:
• ◦ Trained multiple Image classifiers for validating vehicle panels and Mask R-CNN models for damage
• detection on the car panels.
• ◦ Deployed ML models on Triton Inference server, optimising inference with TensorRT for high performance,
• handling peak load of up to 50,000 images per hour, ensuring scalability and reliability in production.
• ◦ Optimising team delivery speed by integrating FiftyOne, open-source tool for dataset curation and analysis,
• into the workflow, boosting the process of analysing and evaluation of the models.
• ◦ Collaborating with annotators, creating task on CVAT, providing them feedback.

Jan 2023 – Jul 2023 · 6 mos · Gurugram, Haryana, India · On-site

• Successfully developed a cutting-edge tool that leverages the power of LLMs (Large Language Models) and HCI
• to analyze and extract valuable insights from unstructured text data, including Employee
• Experience, 360 Feedback, and Candidate Feedback, Remarkable reduction from a 3 month long
• process to mere minutes. Worked independently on R&D side(NLP), Backend and Frontend.

Jun 2022 – Jul 2022 · 1 mo

Feb 2022 – Jul 2022 · 5 mos

• ◦ Built the end-to-end hierarchical Machine Learning classifier to detect the Infographics present in the image that boost the performance of their education recommendation engine for the client( Data Extraction, Experimentation, Training, Deployment).
• ◦ Analyse and Debug the model pipeline and Retraining of the models related to Computer Vision and OCR.
• ◦ Worked on POCs for a client(Document processing and Extraction, NER using Heuristics and ML.)
• ◦ Worked on multi-modal transformer(LayoutLM v3) to fine-tune on NER task.

Aug 2021 – Jul 2022 · 11 mos

May 2021 – Dec 2021 · 7 mos

• Project 1: Automated Optical Coherence Tomography Image Classifier Using a Well Scrutinized Convolutional Neural Network.
• Proposed a CNN network to classify the OCT scan from the Publicly available dataset,
• with minimal image pre-processing, achieving a test value accuracy score over 99%.
• The model is robust is identifying the four primary classes of OCT results: ChoroidalNeoVascularization (CNV), Diabetic Macular Edema (DME), Drusen, and regular eye, which is helpful in diagnosing the human retina. Work includes data preprocessing, image denoising, hyperparameter optimization, model development.
• Project 2 : Deep Learning methods to estimate the torque of a PMSM motor.
• Which includes work on a 74 million time series data, performed the EDA to get the get insights of a data, used PCA to Optimize the time data series in terms of memory usage, Created an artificial neural network to predict the result, proposed 1dcnn to obtain better results.

May 2021 – Jul 2021 · 2 mos

• R&D: Conducting live sessions, posting SMC and articles on OpenCV & GANs.
• Hands-on experience of real time application using OpenCV(image processing, face detection, lane detection) & Generative models(GANs)