Meghana Ganapa | AI, Data, and Software Portfolio

AI Coding Agent using Codex and LLM Workflows

AI-powered coding assistant for code generation, debugging, and developer workflow automation.

AI / LLM

Goal: Automate code generation, debugging, and developer productivity tasks with Codex-backed LLM workflows.
Backend: Built scalable FastAPI endpoints for prompt processing, context management, and AI-assisted workflow execution.
LLM workflow: Implemented prompt engineering, API integration, and response validation to improve code quality and reliability.
Architecture: Applied modular backend design and production-focused automation patterns for maintainability.
Timeline: Developed in May 2026.

PythonCodexFastAPIPrompt Engineering

Code

GridFlex AI - Renewable Energy Optimisation Assistant

Simulated real-time smart grid assistant that recommends energy actions using AWS serverless services and Amazon Bedrock explanations.

AWS AI

Goal: Optimise household energy usage across solar generation, home load, EV charging, battery state, and electricity price signals.
Simulation: EventBridge Scheduler triggers a simulator Lambda that generates telemetry and stores readings in DynamoDB.
Decisioning: A decision Lambda reads the latest telemetry, detects renewable surplus or peak-price demand, and recommends whether to charge the EV, run appliances, charge the home battery, or avoid heavy usage.
Explainability: Amazon Bedrock generates a human-friendly reason for each action, with final recommendations stored in a separate DynamoDB table for traceability and dashboard display.
Future-ready: Designed for integration with smart meters, solar inverters, EV chargers, home batteries, and IoT devices.

AWS LambdaEventBridgeDynamoDBAmazon BedrockServerless

Code Demo

Fraud Claim Detection Agent with AWS Bedrock and Lambda

End-to-end insurance fraud screening agent combining machine learning, serverless deployment, and natural language explanations.

ML Agent

Goal: Screen insurance claims for fraud risk using claim details such as vehicle type, policy type, accident area, police report status, witnesses, and previous claims.
Model: Trained a majority-voting ensemble with Logistic Regression, XGBoost, LightGBM, and CatBoost on the insurance fraud dataset.
Deployment: Packaged the trained model and Lambda handler in Docker, pushed the image to Amazon ECR, and ran predictions through AWS Lambda.
Agent workflow: Connected Amazon Bedrock Agent to a Lambda action group so users can submit claims in natural language and receive a plain-English risk explanation.
Explainability: Returned fraud probability, risk level, recommendation, and warning flags such as no police report, no witness, high claim history, or policy holder at fault.

PythonMLDockerAWS LambdaAmazon ECRAmazon Bedrock

Code

Roo APP 7:03 AM

Today's best AI + startup jobs for Friday, 8 May 2026. Fresh roles with the strongest Australia, remote, AI, and startup fit.

Founding ML Engineer - Remy - Remote
AI/Blockchain Software Architect - Sydney, NSW
Staff Software Engineer - Sydney, NSW

Roo Jobs Daily - AI and Startup Jobs Aggregation Pipeline

Automated pipeline for collecting, ranking, and delivering AI and startup job opportunities.

Automation

Goal: Aggregate, filter, rank, and deliver AI and startup roles relevant to Australia and remote job seekers.
Pipeline: Built Python and FastAPI workflows to collect fresh listings, deduplicate results, and categorise roles by location and job type.
Delivery: Integrated Slack publishing for the top 7 curated jobs every day at 7:00am Australia/Melbourne time.
Quality: Added ranking, filtering, logging, monitoring, and failure-safe alerting to reduce noisy results.
Timeline: Developed in April 2026.

PythonFastAPIPostgreSQLSlack APIPlaywright

Retrieval-Augmented Generation System for Medical Textbooks

Semantic search and LLM question answering over large-scale medical textbook content.

AI / LLM

Goal: Enable AI-powered question answering over large-scale medical textbooks using semantic search and LLMs.
Pipeline: Implemented document chunking, embedding generation, vector similarity search, and grounded response generation.
Retrieval: Used vector databases and CrossEncoder reranking to improve retrieval accuracy and response relevance.
Impact: Reduced manual document search time by approximately 40%.
Timeline: Developed from Apr 2025 to Jun 2025.

PythonSentenceTransformersVector DatabasesCrossEncoder

Code

AI-Driven Polyp Analysis and Colonoscopy Recommendation System

NLP and machine learning pipeline for clinical extraction and follow-up recommendation support.

NLP

Goal: Extract clinically relevant information from unstructured pathology reports.
Models: Used BERT and BiLSTM models to identify polyp size and histology from clinical text.
Engineering: Built custom text parsing and feature extraction for complex clinical dimensional expressions.
Deployment: Deployed Flask APIs to convert unstructured reports into structured healthcare datasets.
Result: Achieved approximately 85% classification accuracy and reduced manual clinical review by approximately 30%.

PythonBERTBiLSTMTensorFlowPyTorch

Early Mortality Prediction in Acute Kidney Injury (AKI)

ML pipeline on ICU data to identify predictors and support clinical insights.

Problem: Predict early mortality risk for AKI patients using ICU data (MIMIC-IV).
Data work: Performed cleaning, imputation, feature engineering, and LASSO feature selection.
Modeling: Trained 7 ML models; best model was XGBoost with AUC = 0.890.
Explainability: Used SHAP + LIME to identify key drivers (e.g., GCS, BUN, urine output, age).
Outcome: Produced interpretable insights suitable for clinical decision support.

PythonSQLXGBoostSHAP/LIME

AI-Powered Legal Assistant

Legal tech platform to analyze case text and predict outcomes (80% accuracy).

Problem: Legal professionals need faster ways to analyse case documents and estimate outcome likelihood.
Approach: Built NLP classification pipeline for case outcome prediction and case analytics.
Data ingestion: Used TypeScript crawlers to collect and preprocess legal text for modelling.
Deployment: AWS-backed platform with dashboards for exploring case insights.
Result: Achieved ~80% accuracy for outcome prediction on legal text.

PythonTypeScriptAWSNLP

AWS Cloud Data Pipeline for ML-Ready Data

Automated ETL to transform raw CSV data into validated Parquet datasets.

Data Eng

Problem: Raw CSVs were inconsistent and not optimised for analytics/ML workflows.
Approach: Built an automated ETL pipeline converting CSV → validated, ML-ready Parquet datasets.
Quality: Implemented schema inference, validation checks, normalization, and partitioning strategies.
Performance: Optimised storage and access via columnar formats to speed downstream feature generation.
Outcome: Improved reliability and usability of datasets for analytics and experimentation.

S3GlueRedshiftParquet

Code

Netflix Analytics Engineering Project

dbt + Snowflake pipeline that transforms raw data into analytics-ready models.

Analytics Eng

Problem: Raw datasets weren’t standardised for reporting and consistent metrics.
Approach: Built analytics engineering pipeline to transform raw tables into curated models.
dbt: Implemented models, tests, snapshots, macros to enforce data quality and historical tracking.
Warehousing: Used Snowflake for scalable analytical queries and a single source of truth design.
Outcome: Improved governance, consistency, and analytics readiness of datasets.

SQLdbtSnowflakeTesting

Code

Sales Performance Analytics Dashboard

Power BI dashboards with star schema modelling and KPI tracking.

Goal: Provide stakeholders a self-serve view of sales performance and KPI trends.
Model: Built star-schema model across products, regions, and customer segments.
Dashboards: Created interactive pages with slicers, drilldowns, trend views, and KPI cards.
Metrics: Tracked Total Revenue (336M) and Sales Quantity (847K) with validated datasets.
Outcome: Improved accessibility of insights for non-technical users and reporting accuracy.

Power BIDAXPower QuerySQL

Code

Insurance Performance Dashboard

Advanced Excel dashboard for policy performance, claims trends, and revenue metrics.

Excel

Goal: Build an end-to-end insurance analytics dashboard without BI tooling (Excel-only).
Analysis: Evaluated policy performance, claim trends, and revenue metrics with KPI summaries.
Techniques: Pivot Tables, XLOOKUP/VLOOKUP, conditional formatting, calculated fields.
Reporting: Enabled dynamic views for ad-hoc analysis and business-friendly reporting.
Outcome: Delivered an easy-to-use reporting solution for business users.

ExcelPivot TablesXLOOKUPReporting

Code

Melbourne Magic: Interactive Tourism Analytics Platform

Interactive maps + dashboards for tourism, transport, and location-based insights.

Dashboards

Goal: Help Victorian travellers explore tourism, transport, and location-based trends interactively.
Data: Integrated multiple public datasets into a single analytics experience.
Build: Developed dashboards with maps, filters, summaries for non-technical users.
Insights: Converted patterns into actionable planning and comparison decisions.
Outcome: Delivered a user-friendly analytics platform for exploration and decision support.

RRShinyTableauData Integration

Hi, I am Meghana. I build AI automation, data pipelines, and backend workflows that turn complex information into useful systems.

Projects

Skills

Experience

Education & Certifications

Contact