AI / ML data product

Real-time crypto data, fused with ML sentiment

KriptoPanel reads market mood alongside price — combining live exchange data with a machine-learning sentiment engine so a signal is never read in isolation.

A product I designed and built end-to-end — solo.
Live inputs
Market · social · news
ML
Output
Live sentiment signal
What it does

Price never read in isolation

KriptoPanel continuously ingests live market data, computes technical-analysis indicators, and scores market mood with a machine-learning sentiment engine — then fuses the two into a single, real-time view. It streams data from exchanges and social/news feeds around the clock, so the signal reflects both what the market is doing and how it feels.

Architecture

How it's built

Streaming ingestion

A real-time pipeline pulling live market data from exchange APIs (Binance, Bybit) alongside social and news feeds — async, always-on.

Technical-analysis engine

Orderflow-aware indicators — Market Profile / Auction Market Theory, CVD, Open Interest, Fibonacci — computed continuously on the live stream.

ML / NLP sentiment engine

An ensemble of transformer and rule-based models scoring social and news text in real time, tuned for crypto and finance language.

Signal fusion

Sentiment score fused with live price into one actionable signal. Production ML serving under real load — not a notebook.

ML / NLP models

An ensemble, not one model

Domain-tuned transformers do the heavy lifting on crypto and finance text, combined with fast rule-based scoring for robustness. A scoring layer on scikit-learn and SciPy turns model output into a usable signal.

CryptoBERTcrypto-domain sentiment
FinBERTfinancial-text sentiment
twitter-robertasocial-post sentiment
VADERrule-based scoring
Tech stack

Under the hood

Python Async streaming pipeline Hugging Face Transformers scikit-learn · SciPy Exchange APIs (Binance · Bybit) NLP / Sentiment analysis PostgreSQL Data engineering