About me
I'm a Data Scientist & AI Engineer with expertise in machine learning, deep learning, and large-scale data systems. My work focuses on building scalable, end-to-end ML solutions that solve real-world problems, from data preprocessing to deployment.
I enjoy working at the intersection of AI research and practical applications, turning complex data into insights and intelligent systems. I aim to contribute to cutting-edge AI innovation—LLMs, generative AI, optimization-driven ML—while driving measurable business impact.
Download ResumeInterests
AI · ML · Deep Learning · Data Engineering · Statistics
Education
M.S. Data Science — Stony Brook · B.E. ECE — Andhra University
Experience
- Scraped and structured 50,000+ housing and social service records across multiple counties into machine-readable datasets.
- Automated web data extraction using Playwright with asynchronous concurrent scraping, reducing collection time by ~70%
- Designed a deduplication framework to merge duplicate organizations while preserving unique attributes, cutting redundancy by ~35%.
- Implemented a relevance-filtering prompt system that improved classification accuracy of housing-related records to >60% precision.
- Delivered a final cleaned dataset for NGO partners, enabling more accurate housing service mapping and supporting advocacy for individuals with serious illness.
- Built Python pipelines to clean and preprocess unstructured data from web pages, PDFs, and other raw formats, version-controlled with Git for reproducibility and collaboration
- Built scalable ETL pipelines in BigQuery and SQL on GCP to support end-to-end ML workflows for anomaly detection.
- Enhanced early fault detection by identifying spikes and irregular patterns in manufacturing time-series data through Z-score thresholds, facilitating effective data visualization for analysis.
- Trained unsupervised models (Isolation Forest, One-Class SVM) to detect anomalies in manufacturing sensor data achieving 78% recall and 73% precision, supporting early fault detection.
- Explained model results to technical and non-technical teams and engaged with data science experts to learn more about the field, supporting fault resolution and alignment.
- Developed real-time demand forecasting and inventory optimization by deploying XGBoost models on GCP using FAST API and Docker, reducing forecast error by 18% across 30+ SKUs.
- Designed end-to-end ML data pipelines on unstructured data and SQL-based ETL workflows using Spark, Hive, and Kafka, accelerating deployment time by 40% and supporting analysis of 1K+ events daily.
- Drove 30% marketing ROI uplift by applying clustering on 10K+ customer profiles, enabling business teams to target high-value segments effectively.
- Conducted A/B testing on promotional strategies and new product placements across multiple regions, identifying winning variants that increased sales conversion by 7%.
Projects
Skills
Open Source & Writing
Open Source Contributions
Haystack Core Integrations (deepset-ai)
Two merged PRs: (1) Fixed llama-stack integration for llama-stack>=0.4.0 — updated api_base_url and CI workflow (PR #2805). (2) Stopped CI from running tests for archived Google integrations — removed workflow files and labeler refs for google-ai-haystack and google-vertex-haystack (PR #2802).
Statsmodels
Contributed documentation improvements (merged PR #9660 — clarified `Gamma` `loglike_obs` and `weights` parameterization) and currently implementing rotated copula support for enhanced tail-dependence modeling.
Skrub
Authored two contributions — one improving documentation and example clarity (merged), and another proposing a new `select_list` / `is_list` selector feature for list-like columns (closed after backend discussions). Both contributions deepened library understanding and informed future selector design.
Outlines (dottxt-ai)
Added parametrized Transformers smoke test for tokenizer robustness (merged PR #1814). Tests initialization and constrained generation across multiple Hugging Face checkpoints to surface tokenizer-related edge cases across models.
Writing & Publications
Prompt engineering was a phase, prompt design is the craft!
Published under AI Mind. Covers the shift from prompt engineering to prompt design as a disciplined creative process.
How Text Chunking Works: The Foundation of Every RAG System
10-minute technical deep dive explaining why chunking strategies determine retrieval quality in RAG pipelines.
How Do You Measure an LLM’s Intelligence? A Complete Guide to Evaluation Strategies
27-minute, 6.6 K-word guide on LLM evaluation metrics — 16 claps and 4 reader highlights.