Mayur Dalal

Leading GenAI initiatives for RnD and multiple domains while handling clients and managing teams. Advises on the strategy and development of multiple products and applications. Leading team size of 18-20, containing Data scientists, Data engineers, MLOps engineers, business analysts and Full stack engineers.

• smart Agent Assist: Created CCaaS solution for multiple clients which results in 10 million $ ARR. Includes transcribing call realtime, various nudges like vulnerability identification, EOD detection, call summary generation using fine tuned LLM (Mistral). Deployed using AWS stack
• smartAudit: Automated calls auditing against 25 parameters using multiple LLMs for volume of 10000 calls daily resulting in saving of 2100 manhours.
• EXL health vertical proprietary LLM: Contributed in creating LLM for health domain.
• KM for contract analysis: Built RAG system for analyzing legal contracts using langchain
• Technology: LLMs, langchain, RAG, AWS (CloudFront, Lambda, EC2, Fargate, ECS, ECR, AppSync, Sagemaker, ELB )

Leading end to end initiatives for cyber security domain

• Command2Vec: Created embeddings from scratch to convert command line instructions to vectors to detect malwares
• Anomaly Detection in VPC logs: Detecting anomalies in ec2 instances packet flows for any unusual behaviour
• SecurGPT: Built security domain RAG system
• Technology: Python, Isolation Forests, Deep learning, MLOps , Git, Docker, MLFlow, DVC, Pytorch, LLM, vLLM

Owned end to end stack for Data, ML and MLOps(CI/CD), actively managing people and product.

• Website/App Search: Developed search for website/app and e-commerce business handling 3 to 4 million requests daily by increasing conversion metrics which resulted in increased revenue (multiple of million dollars) by developing advanced modules like multi-stage query-product retrieval, dynamic reranking of domain based on query affinity, LLM finetuned Help and Support section
• Digitizing Documents: Developed OCR-entity extraction pipelines for various documents like invoices, POI, POA which reduced operational cost by 38%
• Technology: Python, Elasticsearch, Keras, ONNX, TensorRT, git, Docker, CRON, BERT, YOLO, MLFlow, DVC, EvidentlyAI, github actions

• Campaign Activation Problem (Ad CTR optimization): Implemented MABs for ad click for onboarded customers achieving nearly 7% lift over previous approaches
• Hybrid Search: Built hybrid search (keyword + embeddings) for catalog search engine
• Technology: Python, Multi Armed bandits, UCB algorithms, Thomson Sampling, elasticsearch, BERT

• Article recommender: Made recommendation engine for proposing articles based on interaction of users with the articles increasing user engagement by 18% (avg article read time and CTR)
• Risk type Classification: Made risk level classifier as soon as a customer raises an issue
• Entity Extractor from medical journals: Using CRF and BiLSTM extracted entities like diseases, medicines from medical documents
• Technology: Python, keras, Collaboration Filtering, CRF, Bi-LSTM, classification

Predict Service renewal contract for Cisco sells

• Service renewals could potentially account for more than $8 billion annually in recurring revenue.
• Various steps involved understanding features, feature engineering , dimensionality reduction, building various ML models
• Technologies Involved : Python, EDA, Different ML models

Discovery of Latent Topics in Documents

• Created model which could discover underlying hidden topics from documents for support org
• Methods employed were Topic Modelling using LDA (Latent Dirichlet Allocation)
• Technologies Involved : R, Python, Topic Modelling

Priority Prediction for Down Host

• Designed Priority Recommender System for Operations group which usually deals with critical cases using K-NN Approach
• Technologies Involved: K-NN