projects

Adtext icon

Adtext

In Progress

AI chat apps struggle to monetize without harming user experience. Built contextual ad infrastructure that detects conversation intent and surfaces relevant offers natively within chat interfaces.

Dimewise AI Finance APP icon

Dimewise AI Finance APP

Live

Managing personal finances is fragmented across multiple tools and platforms. AI-powered finance dashboard with secure authentication, real-time budgeting, and intelligent insights.

ESP32 AI — LLM on a Microcontroller icon

ESP32 AI — LLM on a Microcontroller

Live

Frontier language models need servers or GPUs, ruling out private, offline inference on tiny hardware. Ported a 28.9M-parameter language model to run fully on-device on an $8 ESP32-S3, streaming tokens to an attached screen at ~9 tokens/second with nothing sent to a server.

AI Voice Receptionist icon

AI Voice Receptionist

Live

A premium wellness resort needed to answer caller questions about packages without adding front-desk load. LiveKit Agents voice receptionist with OpenAI GPT, Cartesia TTS, Deepgram STT, a package-information tool, multimodal turn detection, and noise cancellation.

Radiology Report Harness icon

Radiology Report Harness

New

LLM-generated radiology reports drift from required templates — breaking negation, laterality, and measurements. Template-faithful harness that routes each dictated finding to its field, edits abnormals minimally, preserves normals verbatim, and validates every report; Gemini 2.0 Flash at temperature 0 with deterministic fallback.

Context Policy Platform icon

Context Policy Platform

In Progress

Permissions baked into prompts can't enforce context-dependent access for AI agents calling tools. Policy-first platform that evaluates identity plus live context fresh on every tool call — the same request is allowed or denied based on queue, GPS check-in, and ticket state.

SEO/AEO Blog Engine icon

SEO/AEO Blog Engine

New

AI content agents with write access are unsafe, and most auto-generated posts never match real search demand. Agentic blog engine with a hard invariant — the agent has no write tool. It audits posts, drafts against demand, and posts suggestion cards to Slack; a server handler publishes only after a human click.

Autonomous Insurance Claims Agent icon

Autonomous Insurance Claims Agent

Live

First Notice of Loss intake is manual, slow, and error-prone across PDFs and free-text reports. Backend service that parses claim documents, extracts structured data with OpenRouter and Mistral, validates mandatory fields, routes by priority across 5 workflows, and justifies each decision in plain language.

Empty Cups icon

Empty Cups

Live

Small businesses need a simple way to display listings without a complex CMS. Minimal full-stack app serving listing data via Flask REST API with a dynamic JS frontend, containerized with Docker for easy deployment.

Real Estate Tenant Platform icon

Real Estate Tenant Platform

New

Property managers lack streamlined tools for tenant onboarding and data management. Full-featured tenant management system with property management workflows and secure data handling.

CertifyME icon

CertifyME

Live

Organizations need a fast way to generate and verify digital certificates. Full-stack certification management system for generating, verifying, and managing digital certificates with Flask backend.

Todo App with Authentication icon

Todo App with Authentication

New

Simple task management apps lack secure user-specific data isolation. Todo application with secure authentication, protected routes, and persistent per-user storage.

Firebase Login Authentication icon

Firebase Login Authentication

Live

Implementing secure auth flows from scratch is time-consuming and error-prone. Reusable authentication system with email/password login, user onboarding, and protected routes using Firebase.