FinTech AI Lab

Tools

Data Analyzer

Perform data analysis using advanced machine learning techniques

Earnings Transcript Analyzer

Generate summary and line-by-line sentiment analysis of documents (e.g., speeches, earnings calls)

ERFinData

Extract raw financial data from earnings press releases

file-repo

Lightweight file repository utility for organizing and managing local file collections

KPIx

Perform slide-by-slide analysis and executive summary of a slide deck or KPI report

Macro AI Summary

AI-generated summaries of charts in an multi-sheet excel file

Model Analysis

Drill-down analysis of live excel financial models

Q&A Assistant

Query an unlimited set of pdf files, exhaustively or RAG-based

Speech Writer

Train on a speaker style. Then create speeches in that style just from bullet points

ThinkTank-AI

Create virtual panels or focus groups using AI agents

Open Source

ai-hub

Constellation-style local launch dashboard — click nodes to open URLs, launch apps, run scripts, or open files

chat-TRM

Experimental Tiny Recursion Model (TRM) Chat App with only 3.5M parameters

end2end-desktop-automation

Image-based desktop automation framework for Windows apps — configuration-driven, works with any UI toolkit

image-files-summary

Generate AI-powered summaries of image files in a directory

kaibot

Background AI coach that monitors your Outlook calendar, inbox, and OneDrive — and checks in with short, specific coaching messages throughout the day

oracle-arc-solver

Solver for ARC (Abstraction and Reasoning Corpus) puzzles using oracle-guided search

pdf-files-concatinate

Batch-concatenate multiple PDF files into a single document in a specified order

split-mp4

Split a large MP4 file into n equal-length segments

Writing

Building an ARC-AGI-2 Solver (Part 1)

How a multi-agent Socratic debate architecture was designed to tackle ARC-AGI-2 reasoning puzzles, with agents questioning and challenging each other's solutions.

Building an ARC-AGI-2 Solver (Part 2)

ARC-AGI-2 reveals LLM perception gaps: why vision models struggle to count holes—and how text, high-res images, and reflection improve results

Building an ARC-AGI-2 Solver (Part 3)

Building an ARC-2 solver: from multi-agent debate to a tool-augmented Oracle achieving 50–70%+ on the public evaluation dataset.