• Designed and implemented an end-to-end fraud detection proof of concept (PoC) using knowledge graphs from structured financial data in GCP(VERTEX AI).

• Developed graph-based feature engineering pipelines with NetworkX and GraphFrames for scalable processing.

• Migrated the solution to distributed GraphFrames on GCP Dataproc clusters, enhancing high-volume graph analytics.

• Delivered a production-ready PoC that improved fraud signal quality and model interpretability through graph topology.

• Spearheaded Symplore’s real estate ventures, driving profitable investment strategies and delivering a substantial

revenue boost.

• Implemented complete document classification, analysis, understanding, and extraction, as well as enrichment

pipelines, developing multiple services using Prefect as Data workflow Orchestration

• Engineered and fine-tuned LLMs using LoRA and quantization, enabling precise extraction of critical information

from unstructured data along with RAG techniques and evaluation using RAGAS and custom frameworks

• Deployed and integrated multiple highly scalable Async services to handle batch processing and real-time systems

• Managed Data Science team and mentored junior data scientists towards best practices and proof of concepts

• Developed and fine-tuned multimodal large language models (LLMs) using advanced prompt engineering techniques and RLHF.

• Designed and optimized training datasets to improve model performance across various modalities.

• Conducted thorough evaluations of AI-generated outputs to ensure factual accuracy and reduce hallucinations through feedback loops.

• Led ML team responsible for digitizing and automating biz operations using unstructured data at Phablecare.

• Led Early-Stage chronic Hypertension Data Ingestion for Phablecare on a Peta Bytes scale. Data engineered some data to auto-label data for NER using weak supervision frameworks like Snorkel, skweak and maintained MLOps

• Developed ML pipeline to extract disease-related entities from unstructured text data with coverage of 92% and accuracy of 95%, processing clinical documents worth over 5 Million USD.

• Containerized and deployed all the ML services using AWS Kubernetes and built pipelines to automate flow

* Automated financial data processing and reduced execution time from hours to under 30 seconds

* Built NLP models with 70% precision and limited dataset and developed active feedback loop with F1-score of 0.71

* Developed probabilistic annotations using BERT to minimize manual annotations and trained a SpaCy Custom

NER model to process USD 10 million worth of financial transactions with an active retraining loop.

* Developed an anomaly detection algorithm using PySpark with a Hive backend, processing 10,000 transactions per

week, improving the productivity of the Syndicate Bank finance team by 70% through automation..