Software engineer with hands-on experience building backend systems and machine learning applications end-to-end from REST API design and database integration to model training, ensemble design, and LLM-powered pipelines. Built real-world systems using Python, Java, FastAPI, Flask, PyTorch, TensorFlow, and Groq LLM API, with applied work in computer vision, transfer learning, GradCAM explainability, and automated report generation. Comfortable across the stack backend architecture, AI integration, and frontend wiring.

• I like building things that actually run APIs, pipelines, ML systems
• Comfortable going from model training to deployment and wiring it to a backend
• Curious about secure system design and how things break in production

• REST APIs with FastAPI and Flask
• Deep learning systems — training, ensembling, and serving models
• LLM integrations and automation pipelines
• Backend systems with clean structure and database design

PyTorch FastAPI Groq Llama-4 OpenCV GradCAM

• 4-model ensemble (EfficientNetV2-S, MobileNetV3, DenseNet-201, ConvNeXt-Tiny) for MRI classification
• EfficientNetB4 Attention U-Net segmentation : Dice ~0.88
• Dynamic Risk Index (DRI) with lesion-aware fusion for prediction reliability scoring
• GradCAM explainability + Groq Llama-4 for radiology report generation + PDF export
• FastAPI backend — endpoints for upload, inference, report generation, Google Drive storage

FastAPI SQLAlchemy 2.0 SQLite JWT Alembic

• Single-warehouse inventory system: products, purchase/sales orders, returns, damage write-offs
• Role-based access (Admin/Manager/Employee) with JWT login + revocable refresh tokens
• Immutable stock transaction ledger — every stock change writes one auditable row, never edited/deleted
• Atomic, idempotent order completion (receive/complete are safe to retry)

FastAPI SQLAlchemy (Async) JWT ImageKit

• Async FastAPI backend for photo/video sharing with JWT auth (register/login/verify/reset)
• Media upload and CDN delivery via ImageKit, shared feed with per-post ownership
• Owner-only post deletion, single-origin deployment (API + frontend, no CORS)

TensorFlow EfficientNetV2 Python NumPy

• Converted PE binaries to grayscale image tensors for visual pattern classification
• Fine-tuned EfficientNetV2 on 15K+ samples across 31 malware families
• ~95% test accuracy, macro F1 ~0.96

Languages : Python · Java · SQL

Backend : FastAPI · REST API Design · JWT Authentication · SQLAlchemy · Alembic · Microservices · Asynchronous Programming · Postman

AI / ML : Machine Learning · Deep Learning · TensorFlow · EfficientNetV2 · LLM Integration · Feature Engineering · Model Evaluation · OpenCV · Pandas · NumPy

Databases : MySQL · SQLite · Relational Database Design

Frontend : React.js · HTML · CSS

Cloud & DevOps : AWS · Google Cloud Platform · Git · GitHub

• AWS Certified Solutions Architect – Associate (SAA-C03) Specialization : Packt (July 2026)
• AI Fluency: Framework & Foundations : Anthropic
• Introduction to MCP & Advanced MCP : Anthropic

📧 tharunsridhar@gmail.com 🔗 LinkedIn 🔗 HackerRank 🔗 Hugging Face