Teodin Labs
Data Science Consulting · Financial Services

We turn market noise into
pattern recognition

Teodin Labs helps financial brokerage firms make better client-book decisions through machine learning, market analysis, and real-time decision support.

0Prediction methods
0Microstructure features
0Live tickers tracked
0Global exchanges
K-means client segmentationDTW pattern analysisPPO meta-learningReal-time signal generationRisk pattern identificationVolatility impact assessmentMomentum analyticsDecision support systems
About

Quantitative rigor, practical outcomes

Machine learning applied to market data and client behavior — so brokers can act with confidence when it matters most.

Teodin Labs specializes in data science solutions for financial brokerage firms. We identify patterns in market data and client behavior, helping our clients make informed decisions during periods of volatility and uncertainty.

Our approach combines rigorous quantitative analysis with production-grade implementation — actionable insights that directly improve client-book management and risk assessment.

From clustering client portfolios by risk profile, to meta-learning trading systems that select their own prediction methodology, our work spans research and live deployment.

CORE CAPABILITIES

  • K-means clustering for client segmentation
  • Dynamic Time Warping (DTW) pattern analysis
  • Market event response modeling
  • Volatility impact assessment
  • Risk pattern identification
  • Client behavior analytics
  • Decision support systems
Services

What we build for brokers

Specialized data science services focused on pattern recognition and decision support.

Client Book Analysis

Advanced clustering to segment and analyze client portfolios — identifying risk patterns and behavioral trends that drive decisions during market events.

K-meansRisk ModelingBehavioral Analytics

Market Pattern Recognition

Dynamic Time Warping analysis to find recurring structures in market data and predict client responses to volatility and macro events.

DTWTime SeriesPattern Recognition

Decision Support Systems

Custom analytics platforms delivering actionable insight for client-book management, fusing pattern analysis with live market data.

DashboardsReal-time AnalyticsVisualization
Flagship Research

Medium Frequency Trading System

A meta-learning approach to algorithmic trading — PPO that selects prediction methodologies instead of predicting prices.

No single model wins every regime.

Instead of building yet another price predictor, the system uses Proximal Policy Optimization to learn which of 10 methodologies — from ARIMA and GARCH to LSTM and WaveNet — performs best under current market microstructure conditions.

A dual-head PPO policy consumes a 22-dimensional feature vector of price dynamics, volume patterns, and regime classifications, publishing live signals through Redis with sub-second latency.

ARIMAVARGARCHRandom ForestXGBoostLightGBMBayesianLSTMCNN1DWaveNet
Live signal stream · Redis pub/sub
{ "symbol": "AAPL", "signal": "LONG", "current_price": 150.25, "predicted_price": 151.10, "prediction_method": "XGBoost", "confidence": 0.78 }
MFT system architecture

Full implementation: github.com/egemen-candir/High-Performance-MFT-System
Educational and research purposes.

Projects

Selected work

From algorithmic trading to client analytics — and live tools you can use right now.

PulsePup — AI Infrastructure Momentum

● Live

A live market dashboard tracking the AI infrastructure supply chain — memory, server builders, connectors, PCB, cooling, power, networking. Four-hour rolling momentum plus a cumulative trend of the rolling 4h moving average over one trading week, computed from 1-hour bars across US, Taiwan, Korea, Hong Kong and Shenzhen exchanges.

PythonyfinancePlotlyMomentum Analytics
Launch live dashboard →

Medium Frequency Trading System

Research

Meta-learning trading system using PPO to select optimal prediction methodologies rather than predict prices directly — evaluating 10 methods and adapting to changing market regimes.

PPOPyTorchTimescaleDBRedisIBKR API
View overview →

Client Risk Clustering Platform

Case study

Proprietary K-means system for brokers that segments clients by trading patterns and risk profiles, enabling better portfolio management through volatile periods.

K-meansScikit-learnPostgreSQLFlask

Market Event Response Analysis

Case study

DTW-based tooling that identifies how different client segments respond to market events — predictive insight for client-book management strategy.

DTWTime SeriesPythonPandas
PulsePup · live preview — open full dashboard →
Contact

Let's talk about your client book

Discuss how Teodin Labs can help your brokerage leverage data science for better decisions.

LOCATION

Boston, Massachusetts
United States

EMAIL

info@teodinlabs.com
Financial brokerage consulting
Data science solutions

STACK

K-means · DTW · PPO
Python / R
Machine learning frameworks