Projects

Below are selected projects I’ve built, one by one. Each entry has a short description and a link to the GitHub repo (if public).

VisionSynth

Type: Computer Vision · Object Detection
Status: Completed

Description: VisionSynth is my little experiment in giving videos a pair of eyes. The idea was simple. I wanted a tool that could take any video, run real-time object detection on it, and spit out a clean processed version without messing up the audio.

Highlights:

• Processes videos frame-by-frame using YOLOv8n and keeps a steady ~30 FPS on CPU for most clips.
• Maintains perfect audio synchronization by handling video and audio streams separately in MoviePy.
• Tested on 1,000+ frames to ensure stable detection performance across varied scenes.
• Supports CPU-only execution, making it usable on regular laptops without any GPU dependency.

Code: https://github.com/PiUnknown/VisionSynth

Face Recognition Core Module

Type: Computer Vision · Face Detection
Status: Completed

Description: A minimal real-time face-recognition module that captures webcam frames, detects faces, and identifies them using a lightweight pipeline. No GUI. Just core recognition logic you can plug into any system.

Highlights:

• Captures live webcam frames and performs recognition at ~25–30 FPS.
• Uses a simple classification pipeline built on top of face_recognition to identify faces reliably.
• Evaluated on 2,000+ frames under different lighting and angles to measure accuracy and precision.
• Outputs clear match results while keeping the architecture modular and dependency-light.
• Designed so you can plug it directly into any system (attendance, security, analytics) without rewiring the whole project.

Code: https://github.com/PiUnknown/Face-Recognition-Core-Module

InboxPilot

Type: Experiment · Agentic tooling
Status: Building Phase

Description: A multi-agent system that fetches emails, parses threads, summarizes them using Gemini, assigns priority tags, and stores lightweight memory for personalization. Outputs a fast, actionable digest of every message without opening the inbox.

Highlights:

• Multi-agent pipeline with a Supervisor coordinating Fetcher, Parser, Summarizer and Priority agents.
• Fetches Gmail messages via OAuth and exposes a clean get_emails() tool.
• Parses raw MIME, removes signatures, normalizes text, and expands entire threads
• Summarizer Agent (Gemini) outputs a TL;DR, bullet summary, and concrete action-item list.
• Priority Agent mixes heuristics with LLM reasoning to tag messages as HIGH MEDIUM / LOW.
• Memory Bank stores past summaries, writing preferences, and frequent correspondents for personalization.

Code: *Still cooking