Prajwal Ganugula

About

Software Engineer and ML Systems Engineer

Tempe, Arizona (Open to relocation) | pganugul@asu.edu

I build dependable software and ML systems end to end. My work spans backend services, data pipelines, and applied ML, including computer vision and LLM/RAG systems. I have shipped production systems in Python, C++, and Java, optimized models for edge devices, and published research at ICCV. I focus on clean system design, rigorous evaluation, and reliable delivery.

View Resume View Projects

Experience

Owned a production ML ranking service in Python and C++, exposing gRPC/REST APIs used across internal workflows and increasing straight-through processing of low-risk claims by 18%.
Designed and operated batch and online ML pipelines using Airflow, Spark, and Redis, adding metrics, alerts, and SLOs to maintain latency and error budgets during peak traffic.

Built and deployed recommendation and retrieval services over 100K+ documents, integrating PyTorch models behind APIs and improving click-through rate by 23% via online experiments.
Developed evaluation and experimentation pipelines with dashboards and A/B testing to iterate on ranking quality, feature changes, and user engagement signals.

Optimized on-device LLM inference pipelines for Edge SoCs using FlashAttention, QLoRA, and INT8 quantization, achieving sub-100ms latency under strict memory and power constraints
Delivered production mobile ML features by compressing Stable Diffusion models via distillation and pruning, reducing model size by 50% and accelerating inference by 8x.

Education

Projects

Engineered a zero-shot segmentation pipeline with EdgeSAM + ONNX Runtime, achieving <40ms CPU latency by decoupling the image encoder.
Deployed an interactive demo with Streamlit/OpenCV supporting text prompts and click-based tracking for tool isolation and privacy filtering.

On-device object removal with Edge-SAM selection and diffusion inpainting.

Agentic RAG copilot integrating runbooks, tickets, and live logs.
Built with LangChain/LangGraph, Python, FastAPI, and Docker.

BLIP-2 + ViT-G pipeline fine-tuned with QLoRA and scaled with DeepSpeed.
Multi-GPU training with NCCL and memory-efficient attention.

Built a shared feature store on PostgreSQL and Redis with versioned schemas and validation, enabling consistent online and offline features across ML services.
Developed an experimentation service exposing Python APIs and reusable Airflow jobs, allowing teams to launch ML A/B tests with minimal integration overhead.

Implemented a two-tower ranking model with user and item embeddings and sampled softmax, improving NDCG@10 by 12% over a logistic-regression baseline.
Served models via Python/gRPC services with Redis feature caching and offline training jobs, integrating with an A/B framework to measure engagement impact

Research and Publications

Skills

Languages

Python, C++, Java, SQL, Bash, JavaScript

ML and CV

PyTorch, TensorFlow, Diffusers, TensorRT, ONNX, Segmentation

LLMs and RAG

QLoRA, RAG, LangChain, LangGraph, Evaluation

Platforms

FastAPI, REST, gRPC, Docker, Kubernetes, GitHub Actions

Data and ETL

Airflow, Kafka, Spark, PostgreSQL, Redshift

Cloud

AWS, Azure, Observability, CI/CD

Certificates

View certificate

Awards

View award

Contact

Email: pganugul@asu.edu

Phone: +1 480 942 3759

LinkedIn: linkedin.com/in/prajwal-ganugula

GitHub: github.com/prajwalganugula

About