ALL WORK »

ALL PROJECTS

A collection of projects I've built — from AI voice assistants to automated video pipelines.

AIBACKEND

AI VIDEO EDITOR (V1)

API / CLI

Developed an async Python video editing system with FastAPI + CLI that extracts audio, transcribes with word timestamps, generates AI rough cuts, supports manual segment correction, and exports preview/final outputs plus DaVinci project files.

Problem: Editing talking-head and tutorial videos manually is repetitive and slow, especially when removing retakes, filler words, and pauses before creating a first cut.

Result: Built an end-to-end AI-assisted editing pipeline that automates transcript-based rough cuts, supports human-in-the-loop review, and exports directly to production formats (MP4 and DaVinci-compatible timelines).

PythonFastAPIDeepgramLiteLLM
AIFULL STACK

MEROBOT - PERSONAL AI ASSISTANT

CLI / API

An opinionated personal AI assistant built from scratch with Python, featuring iterative tool-calling, message-bus architecture, and sandboxed execution - enabling complex multi-step tasks through LLM-driven automation.

Problem: Building a robust personal AI assistant requires integrating multiple components: LLM providers, communication channels, tool execution, and maintaining conversation context, all while ensuring security, extensibility, and performance.

Result: Developed a fully functional AI assistant with 8 built-in tools, supporting multiple LLM providers and Telegram integration, demonstrating expertise in async Python development, API design, and agent-based architectures.

Python 3.12+python-telegram-bothttpxBeautifulSoup4
AIBACKENDOTHER

AI SCREENSHOT ORGANIZER

CLI

CLI tool that uses a vision LLM to analyze screenshots, classify them into practical categories, persist metadata in SQLite, and produce a filterable report (grid/table) with Finder shortcuts.

Problem: Large screenshot folders become hard to search and maintain. Manual sorting is repetitive, inconsistent, and usually skips useful metadata like what the screenshot contains.

Result: Built a working AI-powered CLI that auto-categorizes screenshots, renames/moves them into structured folders, stores searchable metadata in SQLite, and generates an interactive HTML report for quick review.

Python 3.12LiteLLMSQLiteLoguru
AIBACKEND

AI TOOLS EXECUTOR

LIBRARY / CLI

A zero-dependency Python library that helps AI agents discover and execute tools efficiently through search_tools, execute, and describe_tool, with safe AST parsing, partial-failure handling, and pluggable search strategies.

Problem: LLM agents often receive full schemas for all available tools on every turn, which bloats context, increases token cost, and hurts tool-selection accuracy.

Result: Built and published a Python package on PyPI that introduces a 3-meta-tool execution layer, reducing tool-context overhead by exposing tools on demand and using Python function-call syntax with AST-based validation. Added robust validation/error formatting and 90 automated tests.

PythonASTPyPIsetuptools
BACKENDOTHER

VISUALKIT

LIBRARY

A pure-Python video composition toolkit that converts JSON timelines into rendered MP4 output using OpenCV + NumPy, supporting media effects, transition pipelines, and animated subtitles for automated short-form content workflows.

Problem: Creating short-form videos from mixed assets (images, clips, subtitles, transitions) usually requires heavy editors or brittle scripts, making repeatable content production slow and hard to automate.

Result: Built and published a reusable Python library on PyPI (v0.1.4) that renders scripted videos through a deterministic JSON project format, with modular effects/transitions, subtitle animations, and configurable resize strategies.

PythonOpenCVNumPySetuptools
FULL STACKBACKENDFRONTEND

OPENALERT

MOBILE APP / API

OpenAlert is a Flutter + Node.js system that turns external events into actionable mobile alerts. It combines JWT-based auth, FCM push delivery, device registration, and alarm-style notification behavior to support faster user response for urgent events.

Problem: Critical alerts can be missed when users rely on passive notifications. OpenAlert was built to deliver high-priority, actionable mobile alerts and connect external events (like inbox webhooks) to immediate user attention.

Result: Delivered a working full-stack alerting system with authenticated mobile users, device token management, webhook-driven push notifications, and in-app alarm controls. The project demonstrates end-to-end ownership across mobile client, API design, authentication, notification delivery, and deployment-ready configuration.

FlutterDartFirebase CoreFirebase Messaging
AIBACKEND

ROXY AI ASSISTANT

CLI / API

Real-time voice AI assistant with full-duplex conversation, dynamic MCP tool usage, and low-latency speech pipeline.

Problem: Voice assistants lack real-time, full-duplex conversation with contextual reasoning and dynamic tool use.

Result: Built a real-time voice assistant with on-device VAD, STT, and fast TTS — enabling natural, low-latency conversations with LLM-powered tool use.

PythonLangChainLangGraphGroq
AIFULL STACK

STORY STREAM AUTOMATION

WEB APP

End-to-end script-to-video automation system with custom render engine, AI generation, and fast multi-resolution export.

Problem: Creating polished video content from text stories requires hours of manual editing and production work.

Result: Engineered a complete AI-powered video automation pipeline, reducing production time from hours to under 2 minutes with 15x faster rendering than MoviePy.

PythonOpenCVNumPyTTS
AIFULL STACK

ORATE ME

WEB APP

Speech coaching platform that evaluates recordings with AI models and delivers contextual feedback, scoring, and improvements.

Problem: People preparing for interviews and presentations have no way to get detailed, AI-driven feedback on their speech quality.

Result: Built a full-stack generative AI platform analyzing 1,000+ speeches with 25%+ better feedback accuracy using Whisper and GPT-4.

Node.jsExpressMongoDBLangChain