Dspyt AI: Uniswap Portfolio Tracker

Proposals Round 5
#1

Summary

Uniswap centered Dapp that utilizes on-chain data to model potential price fluctuations in real-time.

Long Description

Dspytai is the Uniswap powered Dapp that utilizes on-chain data to model potential price fluctuations in real-time.

We combine a browser–based frontend that gives users the ability to determine price averages for any ERC20 contract through the combination of a machine learning model and the Graph.

Dspytai also offers simple, free to use downloadable price prediction data in both text based csv format and image based png and svg formats that are extendable to an NFT.

Current Technical Implementation

Dspytai is built with a combination of React Components and Flask and is hosted on Vercel and Heroku. It utilizes The Graph hosted service to query Uniswap data using Python.

Due to observations of small discrepancies in data, we have focused on reliability of the data feed for the machine learning algorithm and subsequent price predictions.

GitHub mono repository: https://github.com/dspytdao/dspytai

Back End with python: https://github.com/dspytdao/Heroku_Graph_ML

Deliverables:

  • machine learning algorithm implementation converted to vercel lambda
  • utilize > 1 time series and > 1 variables to predict price movements of a coin
4 Likes
#2

Recorded Video: DspytAI: Uniswap Portfolio Tracker. Algovera Round 5 - YouTube
as requested by proposal process Proposal Process | Algovera Docs

3 Likes
#3

I think that there is a possibility of predicting prices within confidence intervals within short time horizons, but previous price does not imply future price, and there are alot of information that are not price history alone, which determines prices. An example might be when a project has insider trading, rugpulls, “irrational exuberance”, project deliverables, or media exposure. Usually the SEC requires this sort of information to be disclosed through the standard company disclosures, but the crypto space does not typically comply with this sort of disclosure requirement, which is part of the reason for the volatility and new legislative attempts and regulatory enforcement actions.

1 Like
#4

In our project we shifted the data 10 steps ahead to generate more of a medium term predictions. Uniswap graphql schemas appear to operate only in part. tokenDayData is the only available schema to obtain the more granular data.

We have generated predictions with LSTM : DSPYT AI | Kaggle

The pipeline with sklearn: dspytai/DSPYT AI Notebook.ipynb at main · dspytdao/dspytai · GitHub

Tested the Flask python api with vercel: GitHub - Pfed-prog/vercel-python . The 50mb limit would allow to load sklearn model. However, vercel api is very convenient for data processing and further integration into the project.

1 Like
#5

We further added YouTUbe video - YouTube
and here is an ethereum price prediction Kaggle Notebook with slightly improved LSTM Pipeline
Ethereum Price Prediction | Kaggle