// hello world 👋
AI / ML Engineer · M.S. Computer Science @ Kent State (GPA 3.97)
I design and ship production AI/ML systems — agentic RAG pipelines, LLM fine-tuning, and serverless model deployment with full MLOps. Published LLM researcher, grounded in strong data science, SQL, and statistical fundamentals.
M.S. Computer Science, Kent State University, December 2025 (GPA 3.97/4.0). Currently working as an AI/ML Intern at Zoo Media and a Volunteer Data Science Research Assistant at Kent State, while contributing to open-source AI/ML projects at Nexus Spring of Code. Previously served as a Graduate Research Assistant at KSU where I published research, built production ML systems, and contributed to peer-reviewed publications. B.S. in Computer Science from JNTU Gurajada, India.
My focus is building and deploying production AI/ML systems — from LLM pipelines and agentic AI to serverless model deployment with end-to-end MLOps. This is grounded in solid data science and analytics fundamentals: predictive modeling, statistical analysis, SQL, and BI dashboards. I'm also a published researcher and active technical writer on Medium & Hashnode.
Enterprise SaaS backend featuring strict PostgreSQL RLS data isolation and an adaptive LangGraph Self-RAG agent with built-in FastMCP server support for Claude and Cursor. Cuts hallucinations by ~85% using local DeBERTa-v3 NLI cross-encoders, backed by 160+ automated tests.
Scalable post-training alignment engine implementing DeepSeek-style GRPO and DPO from scratch in PyTorch. Slashes peak training VRAM by ~50% by eliminating the critic model and integrates a zero-noise SymPy AST verifier achieving 74.8% GSM8K / 49.3% MATH accuracy.
High-throughput Diffusion Transformer (DiT) engine built from scratch in PyTorch using Optimal Transport CFM. Accelerates inference by 22× (down to ~18ms) and reduces memory by 56% via 4-step trajectory distillation and an FP8/INT8/INT4 quantization suite.
Clinical histopathology segmentation pipeline (U-Net++ / ResNet-50) on the PanNuke pan-cancer benchmark achieving 0.92 pixel accuracy. Features Seg-Grad-CAM visual explainability, subgroup fairness audits, and drift tracking aligned with FDA Good Machine Learning Practice (GMLP).
Causal decision engine evaluated across 100K+ real orders (R$15.4M GMV). Combines CUPED variance reduction (compressing A/B test duration by 30–50%), X-Learner uplift modeling for campaign targeting, and LightGBM churn forecasting with SHAP explainability.
Quantitative trading pipeline featuring an Attention-LSTM + Ridge/ARIMA stacked ensemble over 10,000+ ETF sequences, hitting ~75% directional accuracy. Deployed as a serverless ONNX API on AWS Lambda (<100ms latency at ~$0.35/month) with MLflow experiment tracking.
Real-time situational awareness engine clustering 75,000+ incident reports to isolate emergency hotspots. Powered by automated 6-hour ETL ingestion, optimized PostgreSQL indexing for sub-second queries, and a spaCy NLP pipeline geocoding 60% of implicit event locations.
Quantitative risk modeling pipeline analyzing 50,000+ loan records to isolate default patterns. Combines stacked XGBoost and Logistic Regression validated with KS-statistic diagnostics, paired with executive Tableau dashboards driving underwriting decisions.
Production Python libraries published on PyPI. `wranglefy` automates multi-format ingestion, schema inference, and PII masking, while `tprdbreader` provides a high-throughput parser that slashed experimental dataset processing overhead by 80%.
Interactive AI culinary assistant indexing 5,900+ recipes with fuzzy score-based ingredient matching (85% similarity threshold), dietary/allergen filtering, and external Spoonacular API fallback with multi-turn session context.
Published work in LLM architecture surveys and computational translation modeling. View Google Scholar ↗
D. Bhati, T. Le, A. Guercio, C. Auman, D.S. Bandaru · IEEE CAI Conference
📖 Cited 2+D. Bhati, F. Neha, D.S. Bandaru, M. Weber, I.D. Gajera · AI (MDPI) 7(4), 142
📖 Cited 2+D. Bhati, F. Neha, D.S. Bandaru, M. Weber, I.D. Gajera · Preprints
📖 Cited 2+D.S. Bandaru, M. Carl, X. Ren · Translation in Transition 8 (TT8) Conference
M. Carl, T. Mizowaki, A. Ray, M. Yamada, D.S. Bandaru, X. Ren · SKASE Journal of Translation and Interpretation
📖 Cited 2+M. Carl, T. Mizowaki, A. Raj, M. Yamada, D.S. Bandaru, Y. Wei, X. Ren · arXiv preprint
T. Mizowaki, M. Yamada, M. Carl, Y. Wei, A. Ray, D.S. Bandaru, X. Ren · JAITS (26th Conference)
Hand-picked articles most relevant to data analysis, SQL, ML fundamentals, and applied AI.
Open to AI Engineer, ML Engineer, Data Scientist, and Data Analyst roles. Happy to connect!