Samhita P.

Key Skills

Software

Other

AWS SageMaker

Google Cloud Vertex AI

Top Subject Matter

LLM-based prompt optimization

Rag Domain Expertise

Fraud-risk detection for healthcare/Medicaid

Top Data Types

3D Sensor

Text

Image

Top Task Types

Fine-tuning

Data Collection

Prompt + Response Writing (SFT)

Freelancer Overview

Data Scientist at IBM (LLM workflows, generative AI model development and validation). Brings 8+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Other. Education includes Bachelor of Science, Cornell University (2022). AI-training focus includes data types such as Computer Code, Programming, and 3D Sensor and labeling workflows including Fine-tuning and Data Collection.

Labeling Experience

Data Scientist at IBM (LLM workflows, generative AI model development and validation)

OtherFine-tuning

Developed and orchestrated LLM-based analytics workflows involving retrieval-augmented generation (RAG) and prompt chaining to support scalable, consistent automated outputs. Implemented and validated modern generative AI components including autoregressive LLMs with fine-tuned models using LoRA/QLoRA. Produced documentation and executive-ready narratives explaining model behavior and retrieval logic for stakeholders and regulators. • Built Python-based automated pipelines for large-scale Medicaid fraud-risk analytics. • Designed detection/validation frameworks with probabilistic modeling, GAN-inspired architectures, and diffusion-based denoising approaches. • Performed benchmarking and audit-readiness improvements for model-driven projects. • Delivered client-facing POCs and technical training sessions (presentations, Q&A, architecture and inference tradeoffs).

2022 - Present

Undergraduate Research Assistant at Cornell University (sensor data preparation and model training)

Other3D SensorData Collection

Conducted sensor-based machine learning research focused on wearable and wrist-mounted sensor data for behavioral prediction and activity recognition. Built and prepared datasets through data collection, preprocessing, and construction for training, validation, and testing. Trained neural network models to predict face-touching events and full-body movement from sparse sensor inputs. • Engineered time-series features from wearable sensor signals using Python, NumPy, Pandas, and scikit-learn. • Trained Keras neural networks for imminent hand-to-face touching detection and real-time user warning triggers. • Supported dataset creation for 3D body-modeling using wrist-mounted sensors. • Developed deep learning approaches in TensorFlow to infer full-body motion from sparse sensor inputs.

2019 - 2022

Education

Work History