SYSTEMS · PROTOTYPES · TOOLS
A structured record of AI systems I have designed, implemented, evaluated, and documented — from enterprise RAG and agent orchestration to document intelligence.
FLAGSHIP SYSTEMS
Each case exposes the problem, architecture decision, my role, and a measurable result.
Weighted accuracy 98.0% | Average latency reduced to 2.6s
Multi-hospital receipt formats → unified JSON output with end-to-end parsing
1 main flow routes intelligently to 19 subflows (RAG, fact-checking, images, news, and more)
RESEARCH PROTOTYPES
Focused builds used to test emerging interfaces, workflows, and product directions.
Manual meeting notes → one-click structured reports
A conference content processing and report generation platform built with Gemini, BigQuery, and FastAPI, automatically organizing structured trend reports from transcripts, presentations, and scraped web data.
Combine real holdings, realtime quotes, and news context into a structured opening report and decision dashboard so investors can gauge risk and opportunity before the open.
A position-aware AI portfolio assistant: combines holdings, real-time market data, and news to generate opening bell guides and decision dashboards, supporting trend trading and risk assessment.
Realtime voice dialogue · optional Realtime API / Gemini Live
OpenAI Realtime API + Gemini Live + MCP: ChatGPT-like voice assistant with selectable models, supporting real-time voice, streaming responses, dark/light mode, and tool-augmented conversations.
One-stop AI retouching across ID, portrait, travel, and virtual try-on scenes
BloomRender — Let your ideas bloom. An AI-driven professional photo editing and generation studio, powered by Google Gemini API, offering retouching, filters, ID photos, professional portraits, travel photos, thematic photoshoots, duo/group photos, and AI virtual try-on.
Slides + outline → professional video with AI voiceover
Automatically convert slides and Markdown outlines into professional videos with AI voiceovers. Supports PDF/PPTX upload, Gemini script generation, multi-voice TTS, and one-click synthesis and download.
ENGINEERING TOOLS
Smaller tools that turn document, media, and knowledge workflows into repeatable pipelines.
NotebookLM / slide-style PDF → editable PPTX
Convert NotebookLM and general presentation PDFs into editable PowerPoints, combining native PDF parsing, PaddleOCR, page classification, and conditional background reconstruction, balancing editability and visual restoration.
Text → dual-speaker podcast audio in one click
Automatically generate professional podcast audio from text content. Uses AI to convert text into a two-speaker dialogue script, combined with Google Cloud TTS for natural voice synthesis. Supports multiple length modes and real-time progress tracking.
PDF → structured Markdown for RAG and AI preprocessing
Converts PDF into structured Markdown. Features a frontend interface and a FastAPI backend. Employs a hybrid parsing strategy (PyMuPDF + Gemini Vision) and supports RAG and AI parsing preprocessing.
SPEAKING · TECHNICAL EXCHANGE