Rekha Srinivas.

Rekha Srinivas
Chimpiri

Data Scientist building reliable machine learning and generative AI systems.

I am a data scientist with a master's from the University of Maryland. I build and evaluate machine learning and generative AI systems, with a focus on careful experimentation and results I can measure and explain.

Available for Data Scientist and Applied ML roles, and open to relocating.

About

I build machine learning and generative AI systems, and I put as much care into how a model is evaluated as into how it is built.

I have spent 2+ years across data science, applied machine learning, and generative AI, spanning predictive modeling, forecasting, causal inference, retrieval augmented generation, and model evaluation. I am comfortable owning a problem from framing and data preparation through model development and rigorous evaluation on held out data.

My focus is the fundamentals that make results trustworthy: understanding the data, matching the model to the problem, and measuring outcomes with metrics that reflect real performance rather than one convenient number.

Experience

Sep 2025 - Jun 2026

Data Scientist

DreamStudio · Atlanta, GA

  • Built and validated the data foundation for an analytics platform, integrating economic, geographic, and industry data from multiple public sources across 3,000+ US counties, standardizing identifiers and resolving conflicts to produce clean, analysis ready datasets.
  • Trained and benchmarked five model families (Logistic Regression, SVM, MLP, Random Forest, and XGBoost) on a large classification problem, accelerating training on GPU with cuML, correcting class imbalance with SMOTE, and using SHAP to explain the main drivers behind each prediction.
  • Developed a retrieval augmented assistant over technical documentation and code, indexing content into a Pinecone vector store with Voyage AI embeddings and answering natural language questions through Model Context Protocol tools.
  • Built the evaluation workflow that scores LLM responses on retrieval relevance, contextual coverage, answer relevance, and faithfulness, catching retrieval and generation failures and communicating model limitations to the team.
PythonSQLXGBoostcuMLSHAPSMOTEPineconeVoyage AIMCPRAGLLM evaluation
Mar 2026 - Apr 2026

AI Trainer, Data Science and ML Expert

Handshake AI · Contract · Remote

  • Applied human-in-the-loop validation to improve the reliability of AI-generated data science and machine learning solutions, combining expert technical judgment with structured model evaluation workflows.
  • Benchmarked competing LLM responses through pairwise evaluation, scoring outputs across correctness, statistical validity, reasoning quality, instruction adherence, reproducibility, and implementation quality.
  • Diagnosed complex failure modes in model-generated Python, SQL, and ML solutions, including data leakage, invalid cross-validation, inappropriate metric selection, unsupported conclusions, unsafe operations, and missed edge cases.
  • Translated expert analysis into detailed evaluation rationales, corrected reference solutions, and reusable feedback patterns that supported prompt refinement, model-quality improvement, and higher-quality training and evaluation data.
LLM evaluationpairwise evaluationrubric scoringerror analysisPythonSQLprompt refinement
Aug 2024 - May 2025

Graduate Teaching Assistant, Machine Learning Systems

University of Maryland · College Park, MD

  • Reviewed and debugged machine learning systems across training, containerization, and deployment, resolving failures in Docker, Kubernetes, AWS EC2, ClearML, Git, and SQL workflows to get pipelines running reliably.
  • Audited model design, validation strategy, and code quality across 40+ machine learning projects, flagging data leakage, weak validation, and reproducibility gaps and recommending concrete fixes.
  • Standardized reproducible experiment practices, from experiment tracking and environment setup to relational data modeling and SQL, improving the consistency and defensibility of results.
PythonDockerKubernetesAWS EC2ClearMLMLflowGitSQLMLOps
Jun 2022 - Dec 2022

AI/ML Intern, Computer Vision

Coincent · Remote

  • Explored and prepared the dataset before modeling, analyzing class distribution and data quality, correcting imbalance and label issues, and documenting the workflow so results were reproducible.
  • Built an image classification system in PyTorch using convolutional neural networks (CNNs), running structured architecture and augmentation experiments benchmarked against a baseline and reaching over 90 percent test accuracy.
  • Improved model performance with systematic hyperparameter tuning and regularization, validating every change with precision, recall, F1, and confusion matrix error analysis rather than a single accuracy figure.
PythonPyTorchCNNsdata preparationmodel evaluation
Jan 2022 - May 2022

AI/ML Intern, NLP

Internship Studio · Remote

  • Analyzed dataset composition and quality across language pairs with descriptive statistics and error breakdowns, using the findings to prioritize where data collection and cleaning would most improve results.
  • Built the data pipeline for a neural machine translation model, cleaning, tokenizing, and aligning multilingual text, then fine tuning transformer sequence models on the prepared corpus.
  • Benchmarked translation quality against a pretrained baseline with BLEU and per segment error analysis, identifying where additional data and preprocessing most improved low resource language pairs.
PythontransformersNLPdata preparationevaluation

Featured Projects

Faultline, Data Reliability Platform

GitHub

A lineage aware data reliability platform on Snowflake and dbt. It ingests World Bank development data, models it with dbt, and continuously checks it for freshness, schema, duplicate, volume, and value problems. Each incident is traced through the dbt lineage graph to show its downstream blast radius, and a fault injection lab proves the monitoring works by measuring detection recall and false positive rate on controlled failures.

detection recall 100% false positive rate 0% 13 dbt tests

Snowflake, dbt, Python, SQL, Streamlit, GitHub Actions

Agentic RAG with Evaluation Harness

GitHub

A question answering system that returns grounded, cited answers from a large collection of FDA drug labels. It rewrites the query, retrieves with hybrid keyword and semantic search, reranks with a cross encoder, and answers inside a tool using agent loop. A DeepEval test suite and CI gate score every change against a baseline.

contextual recall 0.67 → 0.83 precision 0.82 → 1.00 faithfulness 1.00

Python, LangGraph, hybrid retrieval (BM25 plus embeddings), cross encoder reranking, DeepEval, FastAPI

Real-Time Semantic Segmentation

GitHub

A semantic segmentation pipeline that labels every pixel of a video stream frame by frame. I profiled the full path and found that rendering and frame writing, not the model, were limiting throughput, then optimized those stages to reach real time. The model runs in FP16 on GPU and exports to ONNX for portable inference.

30 FPS real time 9.86 ms per frame 29.82s → real time

PyTorch, Fast-SCNN, CUDA FP16, ONNX, profiling

Causal Inference and Uplift Modeling

GitHub

An uplift model that estimates who changes their behavior because of an advertising intervention, rather than who would have converted regardless. Working from a 14 million row trial, it corrects for confounding with propensity score matching and models the treatment effect with a T-Learner. Results are validated out of sample with the Qini metric and broken down by segment.

avg. treatment effect +0.82 pts 11.9M matched pairs out of sample Qini 0.088

Python, scikit-learn, propensity score matching, T-Learner uplift, Qini evaluation

TaxiCast Demand Forecasting

GitHub

A demand forecasting model that predicts NYC taxi pickups by borough a week ahead with calibrated prediction ranges. It benchmarks five models from a seasonal baseline to a time series foundation model on a rolling origin backtest, keeping accuracy and calibration separate. Probabilistic ranges support planning under uncertainty.

WAPE 13.89% vs 15.80% baseline MASE 0.93 168 hour horizon

Python, Chronos-T5, LightGBM, statsforecast, FastAPI

More projects

Two-Stage Recommender

A movie recommender with a retrieval stage that shortlists candidates and a ranking stage that reorders them with richer signals. Validated on a leakage free time split.

two-tower +18% Recall@20 over ALS

PyTorch, LightGBM, LightGCN, FAISS

View on GitHub

Biomedical Cell Detection

An object detection model that finds and counts overlapping blood cells in microscopy images. Trained a YOLOv8 detector and served it through a small Streamlit application.

mAP50 0.931

Python, YOLOv8, OpenCV, Streamlit

View on GitHub

Customer Churn Prediction

Predicts which telecom customers are likely to leave and surfaces the drivers. Trained a model and served it through a FastAPI endpoint and Streamlit app.

ROC-AUC 0.835

Python, scikit-learn, FastAPI, Docker

Canary-RAG

A defensive layer for RAG systems that plants traceable decoy documents. Attempts to delete, poison, or scrape the index trip a detector and snapshot it for recovery.

3 tripwire detectors

Python, ChromaDB, sentence-transformers, FastAPI

View on GitHub

Real-Time Streaming Analytics

Turns a live event stream into rolling one minute metrics on a dashboard. Built as a Kafka to Flink pipeline packaged in a Docker Compose stack.

1 minute tumbling windows

Kafka, Flink, Elasticsearch, Kibana

View on GitHub

Sitebot, Multimodal Web Assistant

A browser extension that answers questions about a web page using both its text and images. Runs on a serverless AWS backend with multi turn memory.

multimodal, serverless backend

Amazon Bedrock, AWS Lambda, DynamoDB

View on GitHub

Skills

Machine Learning and Statistics

PythonRscikit-learnXGBoostPyTorchregression and classificationclusteringensemble methodscausal inferenceforecastingA/B testinghypothesis testingexperiment designfeature engineeringSHAPSMOTE

Generative AI and Evaluation

RAGLangGraphLlamaIndexvector searchPineconeChromaDBVoyage AI embeddingscross-encoder rerankingModel Context Protocolprompt engineeringDeepEvalLLM evaluationfaithfulness and relevance metricserror analysis

Data Engineering and Backend

SQLPandasNumPyPySparkETL and ELTdata validationfeature pipelinesFastAPIREST APIsKafkaFlinkDuckDB

Cloud, MLOps, and Visualization

AWS (EC2, Lambda, S3)GCPAzureDockerKubernetesGitCI/CDClearMLMLflowONNXStreamlitTableauPower BIMatplotlib

Certifications   Microsoft Certified Azure Data Scientist Associate   ·   Microsoft Certified Azure AI Fundamentals

Education

2023 - 2025

M.S. in Data Science

University of Maryland, College Park

Coursework: Machine Learning, Deep Learning, Natural Language Processing, Big Data Systems, Cloud Computing, Probability and Statistics

2019 - 2023

B.Tech in Electrical and Electronics Engineering

SASTRA Deemed University

Coursework: Machine Learning Techniques, Linear Algebra, Probability, Programming in Python, C, and C++

Get in touch.

I am looking for Data Scientist and applied ML roles, and I am open to relocating. If your team is working on something that fits, I would be glad to hear from you.