Projects & Architectures
A comprehensive catalog of production mobile apps, local LLM orchestration engines, enterprise event-driven cloud pipelines, and academic research publications engineered by Mathews Philip Venattu.
Smart Receipt Scanner & Expense Analytics
Your Billbox
Omnity labs Inc • In Production • App Store & Play Store (v1.0.4)
B2C mobile application built with Flutter. Employs serverless GCP Cloud Functions and the Google Gemini API to parse unstructured paper and digital receipt images into validated JSON records in real time, with event-driven sync to Neon DB PostgreSQL.
- Engineered automated image pre-processing pipelines in GCP Cloud Functions to enhance image contrast and remove camera distortion prior to multimodal LLM ingestion.
- Structured Gemini API prompts with deterministic JSON schema validation to guarantee 100% compliant payload ingestion into PostgreSQL.
- Implemented optimistic local caching in Flutter with background synchronization, ensuring seamless offline receipt captures.
Local LLM Cold Email Generation via n8n
AI Outreach Automation Engine
Omnity Labs Inc • Production Pipeline
Automated cold email generation pipeline connecting Google Sheets to an LLM via n8n. Scrapes company websites, prompts the model for context-aware messaging, and writes personalized drafts back to the sheet. Currently being productized into a public web app.
- Designed an autonomous n8n workflow listening to sheet row updates, extracting domain metadata via web scraping, and generating tailored copy.
- Initially built with a local Ollama inference server running Gemma 7B quantized weights to maintain zero-cost inference and data sovereignty during prototyping.
- Transitioning the architecture into a public web application (omnityspace.com), migrating from local inference to the Gemini API to support multi-tenant scaling and advanced reasoning capabilities.
- Built automated rate limiting, retry backoffs, and duplicate detection directly into the workflow canvas.
Enterprise multi-tenant platform comprising an Admin Command Center, Member Web Dashboard, and native Android application built with Next.js, React, Tailwind CSS, and Firestore, featuring automated subscription billing and role-based access control.
- Architected granular role-based security rules in Firestore isolating multi-tenant administrative, operator, and member domains.
- Built automated subscription billing lifecycles, webhook event processors, and customer renewal alerts.
- Delivered unified design systems and modular UI components shared across the web dashboard and mobile interface.
Planetary Shock Wave ML Classification
Saturn Magnetosphere Classifier
Academic Research • NASA/ESA Cassini Mission
Complex statistical modeling and supervised machine learning analysis on 2005 NASA/ESA Cassini Spacecraft magnetometer datasets to classify bow shock and magnetopause boundary crossings using logistic regression MLE, Random Forest, and Lasso.
- Applied Maximum Likelihood Estimation (MLE) logistic regression to evaluate predictor significance and eliminate non-informative variables.
- Trained and tuned Random Forest and Lasso models with cross-validation in R using caret and ranger packages.
- Achieved elevated sensitivity metrics for classifying rare high-speed boundary transitions in space physics magnetometer data.
Inductive Power Transfer Architectural Simulation
Dynamic EV Wireless Charging Scheme
Peer-Reviewed Publication • Taylor & Francis / CRC Press
Co-authored peer-reviewed research chapter in Emerging Technologies for Sustainability exploring electromagnetic coil design, power electronics, and dynamic inductive power transfer topologies for in-motion electric vehicle charging infrastructure.
- Simulated resonant magnetic coupling architectures between primary track transmitter coils and secondary vehicle receiver coils.
- Evaluated power transfer efficiency and electromagnetic field distributions under lateral vehicle misalignments.
- Chapter published in Taylor & Francis / CRC Press book series on Emerging Technologies for Sustainability.
Need a tailored architectural prototype or data system?
From 0-to-1 full-stack shipping to cloud telemetry pipelines and local AI automation, let's build production-grade solutions together.