Prajwal Mani

Data Scientist

New York City, United States4 yrs 2 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Increased revenue by 25% with dynamic pricing.
Reduced processing time by 40% through AWS optimization.
Built scalable NLP automation for enterprise workflows.

Stackforce AI infers this person is a Data Scientist specializing in Machine Learning and AI solutions for Fintech and SaaS industries.

Contact

Skills

Core Skills

Machine LearningData Science

Other Skills

Data IntelligencePython (Programming Language)LangChainJina AICross-EncoderContextual BanditAWSHyperoptSQSSageMakerKubernetesAirflowPythonPandasArtificial Intelligence (AI)

About

I ship AI that drives results - dynamic pricing, RAG search and chat, and production machine learning. Recent wins: +25% revenue, +17% conversion, -40% processing time. Stack: Python, PyTorch, AWS (SageMaker), Docker, Kubernetes. I focus on reliability, cost, and fast iteration with product and engineering. Highlights • Increased revenue 25% with dynamic pricing • Improved conversion 17% with A/B tests and personalization • Reduced processing time 40% by optimizing AWS pipelines • Built scalable NLP automation for enterprise document workflows What I work with • Machine Learning: deep learning, NLP, transfer learning • MLOps and Cloud: AWS (SageMaker, EC2, S3), Docker, Kubernetes, Terraform • Data and Code: SQL, PySpark, Python, plus Java, C++, R, JavaScript

Experience

4 yrs 2 mos

Total Experience

1 yr 6 mos

Average Tenure

1 yr 1 mo

Current Experience

Forman technology

Data Scientist

Apr 2025 – Present · 1 yr 1 mo · Remote

Cut manual underwriting processing by 70% for 180k+ docs by deploying a YOLOv8 and GPT-4 Vision pipeline, engineered custom logic to parse complex financial tables from unstructured tax returns and balance sheets.
Boosted retrieval accuracy to 82% (Recall@20) for credit risk analysis by architecting a two-stage LangChain RAG pipeline with Jina AI embeddings and Cross-Encoder reranking.
Reduced GPU costs by 55% via TensorRT-LLM compilation and vLLM orchestration, achieving 3x throughput on NVIDIA A10G instances through optimized KV-caching
Increased throughput 40% by routing 80% of traffic to low-latency vLLM endpoints via a semantic routing agent
Established telemetry and drift detection to monitor performance and support monthly retrains, integrated automated logging to ensure production stability in a dynamic environment.

Data IntelligencePython (Programming Language)Machine LearningData Science

Clouddata technology llc

Data Scientist

Aug 2023 – Feb 2025 · 1 yr 6 mos · Remote

Boosted Domestic ancillary revenue by 25% across 500k+ transactions for the Merchandising unit by deploying a Contextual Bandit pricing engine for baggage and services, outperforming legacy static-rule baselines.
Engineered a Thompson Sampling MAB framework on AWS to automate dynamic traffic for 100k+ users, optimized exploration-exploitation to minimize regret and accelerate reward convergence.
Improved AUC to 0.85 and cut retraining time 75% via SageMaker Pipelines with Hyperopt for Bayesian tuning.

Python (Programming Language)Machine LearningData Science

New jersey institute of technology

Graduate Research Assistant

May 2022 – May 2023 · 1 yr · Newark, New Jersey, United States

Engineered a 1.5TB/day NASA pipeline using SQS and SageMaker to process 3D solar data into multi-modal image and tabular datasets.
Developed a PyTorch ConvLSTM-Attention model on 5TB of data, improving eruption prediction accuracy by 18%.
Standardized MLOps via DVC and MLflow to ensure reproducibility, reducing engineering ramp-up time by 50%.

Python (Programming Language)Machine LearningData Science

Intellisense software private limited

Machine Learning Engineer

Jan 2020 – Aug 2021 · 1 yr 7 mos · Bengaluru, Karnataka, India · Remote

Automated high-scale ETL for Transformer models using PySpark on AWS EMR and Airflow, ingested 700GB+ of weekly unstructured text with optimized partitioning, increasing velocity by 40%.
Architected an AI inference service on Kubernetes (EKS) with Horizontal Pod Autoscaling (HPA), maintained 99.9% uptime and low-latency performance at a peak throughput of 25k+ requests/hr.
Deployed BERTopic and Transformer-based summarization models via AWS API Gateway and Lambda, achieving an 80% reduction in manual effort for 500+ weekly assets.

Verzeo

Machine Learning Intern

Jan 2020 – Mar 2020 · 2 mos

Designed a fraud detection system with 88% accuracy, reducing fraudulent activity by 20%, which safeguarded approximately $100K in potential revenue losses for a financial services client.
Partnered with cross-functional teams to analyze and prepare datasets using Python and Pandas, streamlining workflows and accelerating machine learning model deployment for faster project delivery.

Machine LearningData Science