// ML ENGINEER PIPELINE ACTIVE

Building the Next Gen of
Cognitive Systems

Specializing in explainable NLP, multimodal fusion, and medical computer vision. Bridging the gap between raw data and clinical/logistics solutions.

SYSTEM_MONITOR.sh
> python -m model.train --epoch 100
[INFO] Initializing PyTorch CUDA Core...
[SUCCESS] GPU Accelerated: RTX 4090 Active
Training Progress:
98.4%
Accuracy
0.034
Loss
6+
Demos

The Engineer Behind the Algorithms

I hold a Master of Science in Data Science from Saint Peter's University and a Bachelor of Technology in Computer Science & Engineering. My focus lies in architecting robust, explainable ML models that handle complex multimodal datasets.

From enhancing diagnostic pipeline explanations using Grad-CAM to building Reciprocal Rank Fusion engines for retrieval-augmented generation (RAG), I construct systems that emphasize transparency and production scalability.

M.S. Data Science

Saint Peter's University (2026)

B.Tech CSE

GITAM University (2023)

Technical Skill Matrix

Languages & Libraries

Python PyTorch TensorFlow Scikit-learn SQL NumPy Pandas

Deep Learning & NLP

Transformers Hugging Face BERT DeBERTa RAG Explainable AI

Computer Vision & Graphs

OpenCV Grad-CAM NetworkX CNN Image Augmentation

Virtual ML Playgrounds

Experience my machine learning implementations running live in your browser. Interact with inputs and see real-time predictions.

ArgRAG: Argument-Aware Retrieval-Augmented Generation

My research paper project that inserts an argument filtering layer between retrieval and generation, improving factuality by 3.4% and explanation quality by 21%.

1. Retrieve
BM25 (Sparse)
DPR (Dense)
2. RRF Ensemble
Reciprocal Rank Fusion
3. DeBERTa Classify
SUPPORT / ATTACK
4. Generate
FLAN-T5 LLM
ArgRAG_OUTPUT_STREAM
// Click "Run RAG Pipeline" to execute the pipeline flow...

Multimodal Clinical Outcome Prediction (MIMIC-IV)

Integrates structured EHR patient statistics and BioClinicalBERT clinical note embeddings. Adjust the sliders and input patient keywords to observe the fusion model risk calculation.

15% Mortality Risk
STATUS: CLINI_STABLE
// BioClinicalBERT Embedding
Embedding: [0.012, -0.452, 0.912, 0.088, ...]

Anomaly Detection & Grad-CAM Interpretability

Visualizing deep neural network decision features. Toggle the conditions and slide the opacity selector to see how Grad-CAM localizes pulmonary pathological features on the radiograph.

Neural Activation Map

Target Class: Pneumonia
Model Confidence: 94.2%
Grad-CAM Extrapolator: CONVERGING
Findings Summary:

Model localizes high activation intensities in the bilateral lower lung zones, consistent with bronchopneumonic infiltrates.

Graph Route Optimizer with Real-Time Constraints

Calculates optimal delivery routes using graph search algorithms. Click on the grid map to place custom delivery addresses/nodes, select your solver, and trigger the search path optimization.

Click grid to add path nodes

Path Metrics

Nodes Traversed: 0
Total Distance: 0 px
Search Frontier: Idle
Graph Explorer Trace:
  • Waiting for node placement...

Professional Timeline

2025 - 2026 Saint Peter's University

Graduate Research Project

Architecting Advanced NLP Retrieval Solutions

  • Designed and deployed ArgRAG, a modular RAG pipeline adding argument filters between BM25/DPR ensembled retrievers and generators.
  • Optimized classification models (RoBERTa, DeBERTa) to filter passages by argument alignment, boosting factual accuracy by 3.4%.
PyTorch Transformers HuggingFace DeBERTa
2025 Clinical EHR Project

Multimodal Deep Learning Architect

Deep Fusion for Clinical Predictions

  • Built deep fusion architectures combining structured tabular hospital statistics and BioClinicalBERT clinical note embeddings.
  • Improved mortality and triage prediction F1 metric by 7% using PyTorch deep neural networks.
PyTorch SQLAlchemy MIMIC-IV BERT
2024 NASA Dataset Research

MLP Neural Net Architect

Asteroid Diameter Regression

  • Optimized a multi-layer perceptron neural network using TensorFlow/Keras on NASA's open datasets, estimating asteroid size with high $R^2$ scores.
TensorFlow Keras Scikit-learn Pandas

ML Terminal Interface

Run commands directly against my digital twin. Query skills, fetch resume details, or trigger model training simulations.

sta@neuroshell:~
CORE: ACTIVE // LATENCY: 24ms
SURYA TEJA ANUPINDI - COGNITIVE SHELL v1.0.0
Type help to list active terminal utilities.
 
sta@neuroshell:~$

Initialize Connection

> [SUCCESS] Transmission successfully delivered to Surya's node. Thank you!