Skip to content
View ranjithguggilla's full-sized avatar
💭
Building AI systems and research-driven software
💭
Building AI systems and research-driven software
7 followers · 12 following

Block or report ranjithguggilla

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ranjithguggilla/README.md

Hi, I'm Ranjith Guggilla

Gen AI / ML Engineer · building LLM applications, RAG systems, and production machine learning pipelines on Azure and AWS.

I build Retrieval-Augmented Generation apps with Azure OpenAI, LangChain, and Azure AI Search, ship ML models behind FastAPI services, and care about the parts that make AI trustworthy in production: grounding, guardrails, evaluation, and monitoring.

Python Azure OpenAI AWS Bedrock LangChain FastAPI scikit-learn Docker

What I work on

  • Generative AI: RAG over enterprise documents with Azure OpenAI, LangChain, LangGraph, Prompt Flow, and Azure AI Search; prompt engineering with grounding, conversation memory, and guardrails that cut hallucinations by 20%.
  • Document AI: ingestion pipelines with Azure AI Document Intelligence, OCR, and chunking for semantic and vector search.
  • Machine learning: supervised models for risk, fraud, segmentation, and forecasting with scikit-learn, XGBoost, and Amazon SageMaker, tracked with MLflow.
  • Production: FastAPI and Flask inference APIs, Docker, Azure App Service, AWS Lambda, CI/CD with GitHub Actions and Azure DevOps, plus monitoring in Azure Monitor, CloudWatch, and Power BI.

Featured ML projects

churn-radar

Customer churn early-warning system. Compares logistic regression, random forest, and XGBoost (ROC-AUC 0.845), reaches 50% of churners by calling the top 20% of risk scores, and serves predictions with plain-language risk drivers through a FastAPI service.

FastAPI XGBoost Docker MLflow CI

riskbeacon

Credit-risk early warning for two-year financial distress. Class-weighted CatBoost beats a logistic baseline (test ROC-AUC 0.867), and the approve/decline threshold minimizes expected loss in dollars instead of defaulting to 0.5.

CatBoost imbalanced-learn Gradio CI

forgewatch

Predictive maintenance for a CNC milling cell. A glass-box failure model, conformal triage bands with a measured coverage guarantee, survival-based tool-life budgets, and cost-optimal alert thresholds.

Explainable AI MAPIE lifelines CI

pedalcast

Hourly bike-share demand forecasting. LightGBM reaches R² 0.91 and cuts mean error from 102 to 43 bikes per hour against the seasonal baseline, validated on a strictly chronological hold-out.

LightGBM Polars Streamlit CI

curecast

Concrete compressive-strength regression (R² 0.93), run in reverse to find the lowest-carbon mix that still meets a target strength.

scikit-learn Optimization CI

fieldwise

Crop recommendation engine (99.5% held-out accuracy) with an advisory layer that explains soil-nutrient shortfalls in plain language. Optuna tuning and Pandera data validation.

Optuna Pandera CI

EDITH-voice-assistant

Python conversational voice assistant with speech recognition, text-to-speech, and NLP-driven intent handling for weather, news, Wikipedia, messaging, translation, and system control.

NLP Speech
Data engineering and scientific data projects

Tech stack

Category Tools
Generative AI Azure OpenAI, Azure AI Studio, Azure AI Search, LangChain, LangGraph, Prompt Flow, RAG, prompt engineering, function calling, embeddings, AI agents
Machine learning scikit-learn, XGBoost, TensorFlow, PyTorch, MLflow, feature engineering, hyperparameter tuning, model evaluation
NLP and document AI Azure AI Document Intelligence, OCR, tokenization, sentiment analysis, entity recognition, summarization
Vector databases FAISS, ChromaDB, Azure AI Search vector index
Languages Python, SQL, Java, JavaScript
Frameworks FastAPI, Flask, Streamlit, Gradio
AWS SageMaker, Bedrock, Lambda, EC2, S3, IAM, CloudWatch, Glue
Azure OpenAI, AI Search, Machine Learning, Functions, Blob Storage, SQL Database, Key Vault, App Service, Monitor
Data engineering PySpark, Pandas, NumPy, ETL, SQL Server, PostgreSQL, MySQL, MongoDB
MLOps and DevOps Docker, GitHub Actions, Azure DevOps, CI/CD, pytest, Git
Visualization Power BI, Matplotlib, Plotly

Certifications

  • AWS Cloud Foundations
  • AWS Machine Learning Foundations
  • PCAP: Certified Associate in Python Programming
  • MySQL Basics

Education

  • M.S. Computer Science, Texas A&M University–Corpus Christi
  • B.Tech. Computer Science and Engineering (AI & ML), Vaagdevi College of Engineering

Contact

Email LinkedIn GitHub

Pinned Loading

  1. EDITH-voice-assistant EDITH-voice-assistant Public

    EDITH – A Python-based virtual voice assistant with speech recognition, text-to-speech, weather, news, Wikipedia, WhatsApp, email, YouTube, translation, and system control. Packaged with CI, docs, …

    Python

  2. churn-radar churn-radar Public

    ChurnRadar: customer churn early-warning system. XGBoost vs logistic regression, business-impact targeting, FastAPI scoring API, Docker, Gradio UI, MLflow tracking

    Python

  3. curecast curecast Public

    Concrete strength regression (R² 0.93) plus inverse search for the lowest-carbon mix that hits a target strength. scikit-learn, Gradio, CI.

    Python

  4. fieldwise fieldwise Public

    Crop recommendation engine (99.5% accuracy) with a plain-language soil-nutrient advisory layer. scikit-learn, Optuna tuning, Pandera validation, Gradio.

    HTML

  5. forgewatch forgewatch Public

    Predictive maintenance for a CNC cell: glass-box failure model, conformal triage bands (MAPIE), survival-based tool-life budgets, cost-optimal thresholds.

    Python

  6. pedalcast pedalcast Public

    Hourly bike-share demand forecasting with LightGBM (R² 0.91, MAE 43 bikes/hr vs 102 for the seasonal baseline). Time-aware validation, Streamlit dashboard.

    Python