May 2024 – Dec 2024 · 7 mos · Bengaluru, Karnataka, India · On-site

Mar 2023 – Apr 2024 · 1 yr 1 mo · Mumbai, Maharashtra, India · On-site

• GenAI - Audio Analytics for Debt Management :
• Experimented with AWS Transcribe & Open AI Whisper for speech to text and Nvidia NeMo & pyannote
• for speaker diarisation for multilingual mono channel collections agents calls
• Extracted features from the transcribed text using langchain and the LLM - GPT 3.5 turbo
• Ran experiments for hierarchical prompt engineering, Language Identification (LID) & post-processing
• techniques for quality measurement
• Developed a RAG application to retrieve relevant information from bank documents in response to inquiries made during customer calls. Tested Pinecone and Chroma DB for the vector database.
• Developed a credit risk scorecard utilizing XGBoost to determine eligibility for Flipkart EMI using Bureau footprint in Pyspark (Tech stack : AWS EMR and Sagemaker)
• Used Isolation Forest for Anomaly Detection to identify fraudulent loan applications from clickstream data

Jun 2022 – Jan 2023 · 7 mos

• Sales Process Intelligence
• Incorporated descriptive & predictive analytics within sales process to optimise lead targeting & agent performance
• Segmented the sales workforce based on input and output metrics to identify process inefficiencies; worked across
• verticals and stakeholders to rework processes that achieved 65% improvement in productivity
• Transcribed Sales demos & calls using OpenAI Whisper to arrive at a targeted sales pitch using scoring metrics
• Text Analytics Engine for Grievance Redressal
• Classified grievances & redirected them to relevant business verticals, such as tech support, platform, counsellor and doubt-solving; achieved 0.75 macro-F1 over ten classes
• Decreased the TAT for grievance resolution by facilitating quick follow-up over call, email or whatsApp
• Text to Talking Face Video Generation
• Recreated synthetic media of existing educators for content generation from text and image generative AI techniques
• Aligned phonemes from audio to text using P2FA and calculated face key-points from extracted video frames using
• OpenPose to map phonemes to poses
• Generated video from interpolated phoneme poses using vid2vid GAN
• Further improved the lip sync with the TTS generated audio using Wav2Lip for a more natural output
• Vnnovate Hackathon by Vedantu (1st Prize)
• Presented an NLP module for automating question/quiz generation with answers from any academic document
• Extracted keyphrases, paired them up with respective contexts and generated context-aware question-answer pairs leveraging BERT transformer

Dec 2019 – Jun 2022 · 2 yrs 6 mos

• In-class Personalised Audio Nudges to Students
• Bootstrapped custom voice engines from first-party transcribed audio data with intensive audio regularization
• E2E trained and fine-tuned tacotron2 to optimize TTS outputs for Indian English vocabulary and accents
• Stitched pre-recorded & generated audio outputs to handle performance degradation due to out-of-vocabulary
• proper nouns
• Automated teaching session scoring based on multimodal features
• Created a scoring framework that rates in-session teacher presence to facilitate quality audits
• Extracted and scored video features such as facial sentiment analysis, head movement, distance-from-camera and
• brightness; audio features such as pitch variation, speech rate and audio emotion
• Harnessed multimodality by combining features using boosted ensemble classifiers
• Mentored an intern on generating BERT phrase embeddings to find spans that answer textual audit questions
• Lead Scoring
• Evaluated likelihood of users to convert as paid subscribers based on user intelligence synthesized from platform activity, demographic and behavioural information
• Extensively preprocessed and engineered features and tailormade the training regimen, model architecture and model complexity to achieve robust generalization
• Executed an evaluation, prediction and lead assignment pipeline that achieved 50% reduction in daily lead volumes thereby reducing business burn, boosting sales team efficiency by over 10%
• Won first prize in the Vnnovate (hackathon)
• Presented an NLP module for automating question/quiz generation with answers from any academic document
• Extracted keyphrases, paired them up with respective contexts and generated context-aware question-answer pairs leveraging BERT transformer

Jun 2019 – Dec 2019 · 6 mos

• Automated and simplified the pipeline creation process for multiple Lines Of Business
• Tested an inner source tool ’Quantum’ built on Spark to create new components
• Replicated existing pipelines in production end to end using Quantum across financial modules

May 2018 – Jul 2018 · 2 mos · Bangalore, India

• I received a Pre Placement Offer from the company at the end of the internship period.
• Designed an efficient realtime streaming engine with live metric calculation and real time reporting on Grafana
• Tested streaming engines like Spark Streaming and Kafka Streams for processing data taken from Kafka Queue
• Used InfluxDB as a Time Series Database for efficient querying and storage of data
• Experimented with various approximation/streaming algorithms for efficient and quick calculation of metrics

May 2017 – Jul 2017 · 2 mos · Mumbai Area, India

• Designed an application using Android Studio to streamline the survey-form filling process
• Integrated a django based backend admin panel to create, edit and send a survey form to the app