Srinivas V. Koduru, PhD, AI/ML

// AI Tooling Practice

Agentic AI & LLM-Assisted Development

Daily builder — Claude Code, Codex, LangChain, AutoGen, CrewAI, RAG

⚡ Claude Code builder Codex pair-programming LangChain · AutoGen · CrewAI RAG · Vector DBs MLOps · Locked-ML

I am an active daily builder of agentic AI solutions with Claude Code and Codex-style LLM-assisted development, shipping production artifacts into regulated clinical environments. Through this practice I have delivered Flask scientific web applications for diagnostic reporting and end-to-end ML pipelines (random forest, SVM, deep neural networks with SHAP interpretability) integrated into FDA / CLIA-grade automated clinical reporting. I build RAG-based literature workflows over PubMed, UniProt, TCGA, GEO, and ChEMBL, and agentic AI workflows on LangChain, AutoGen, and CrewAI for hypothesis generation, target prioritization, market-opportunity identification, and experimental-design support. I also automate regulatory-document generation aligned to FDA 510(k) preparation and ship data-visualization dashboards to clinical and research stakeholders. The convergence of life-sciences AI/ML, regulatory rigor (CLIA / CAP / NYSDOH / 21 CFR Part 820 / IVDR), and agentic development methodology is the operating model I bring to every pharma R&D, biotech, or precision-medicine engagement.

Flask Diagnostic Dashboards

Production-grade Flask web applications surfacing locked-ML classifier outputs (classification score, confidence interval, actionable recommendation) per patient sample, integrated with CLIA / CAP-compliant chain-of-custody.

Locked-ML Classifier Deployment

Frozen / locked ML models (random forest, SVM, deep neural networks, SHAP interpretability) deployed under FDA / CLIA-regulated automated clinical reporting infrastructure. The highest-bar version of MLOps in healthcare.

Agentic AI Workflows

LangChain, AutoGen, and CrewAI agents built for hypothesis generation, literature synthesis, target prioritization, and experimental-design support — reasoning, planning, tool execution, and iterative refinement in scientific domains.

RAG over Scientific Databases

Vector-embedding + retrieval pipelines over PubMed, UniProt, TCGA, GEO, ENCODE, ChEMBL. Agent skills designed for molecular biology, genomics, biomarker discovery, and competitive-intel research.

AI-Assisted Regulatory Documentation

Automated drafting and assembly of regulatory documentation packages for FDA 510(k), CLIA, CAP, and NYSDOH CLEP submissions — SOPs, validation protocols, analytical / clinical performance documentation, QMS documents.

Codex Pair-Programming Practice

LLM-assisted development used to translate research-grade workflows into production-grade pipelines — Python, R, Nextflow, AWS, HPC, Docker. A core part of my daily build process.

Claude Code Codex LangChain AutoGen CrewAI RAG / Vector DBs Flask PyTorch TensorFlow scikit-learn SHAP Python R / Bioconductor Nextflow AWS Docker Git / GitHub MLOps Locked-ML FDA 510(k) CLIA / CAP / NYSDOH 21 CFR Part 820 ISO 13485 CLSI EP05/06/07/17

// AI/ML Showcase

Cancer Prediction Models

Transcriptomics & Advanced Data Analysis

Decision Tree — Gene-Based Cancer Classification

Deep Neural Network — Multi-Layer Classifier

AI/ML Pipeline Architecture

Per-Sample Classification Dashboard

ROC/AUC with 95% CI

AUC > 0.99 demonstrates near-perfect diagnostic separation between cancer and control samples.

2D ivis Embedding

Clear cluster separation confirms distinct molecular signatures between disease states.

Multimetric Radar — Model Comparison

SVM achieves balanced performance across all metrics, outperforming ensemble and regression models.

SHAP Waterfall Plot

SHAP values reveal which genes drive each prediction, ensuring clinical interpretability.

Threshold — Sensitivity Analysis

Optimal threshold maximizes sensitivity while maintaining >99% specificity for clinical use.

Model Calibration Curve

Well-calibrated predictions ensure reported confidence scores align with actual outcomes.

LOOCV Performance Comparison

Confusion Matrix — Cancer Test Model

Gene Expression Profiles Across Cancer Types

Gene Expression Bubble Matrix — Multi-Cancer Panel

Low

High

Oncogene Tumor Suppressor Signaling

Research Domain Distribution

What's New

From Bench to Bedside — The Story

Research Experience

Leadership Impact

Currently Working Projects

Agentic AI & LLM-Assisted Development

Cancer Prediction Models

AI/ML Pipeline Architecture

Per-Sample Classification Dashboard

Gene Expression Bubble Matrix — Multi-Cancer Panel

Technical Skills

Peer-reviewed Publications

Peer-Reviewed Abstracts

Patents

Academic Background

Get In Touch

Let's Collaborate

Visitors Worldwide