Exchlytics AI

Inspiration

In financial trading, milliseconds matter. Yet analyzing packet capture (PCAP) data from trading systems is complex, manual, and time-consuming. We were inspired to build a tool that empowers analysts, brokers, and regulators to understand and visualize trading patterns, network delays, and anomalies in real time. Our goal was to merge AI-driven insights with network-level transparency, helping detect unfair trades, latency bottlenecks, or potential fraud before they impact the market.

What it does

Exchlytics AI is an AI-powered PCAP analysis platform designed for high-frequency trading environments.
It automatically:

Parses PCAP files using Pyshark and Scapy
Identifies trade-related packets and extracts metadata like order IDs, timestamps, and message types
Uses LLMs (Ollama + Azure AI Foundry) for natural-language analysis of trading patterns
Visualizes latency trends, order flows, and anomalies on an interactive Streamlit dashboard
Generates AI-driven summaries explaining unusual trading activity or network behavior in plain English

How we built it

We combined modern AI and networking tools into a unified pipeline:

Data Parsing: Used Pyshark and Scapy to extract features (source/destination, timestamps, trade message flags) from PCAP files
AI Analysis: Integrated Ollama LLMs for prompt-based reasoning and Azure AI Foundry for scalable inference
Visualization: Built an intuitive Streamlit dashboard for timeline visualization, latency heatmaps, and trend detection
Backend: Python-based modular architecture ensuring easy integration with future ML or time-series forecasting modules
Deployment: Containerized using Docker, enabling quick cloud deployment or local execution

Challenges we ran into

Parsing complex PCAP data since financial packets often use proprietary formats, mapping them to meaningful trade events required custom parsers
Ensuring accurate microsecond-level timestamp comparisons between packets
Getting LLMs to produce domain-specific summaries through refined prompt engineering
Handling large PCAPs efficiently using asynchronous data streaming and memory optimization

Accomplishments that we're proud of

Built a fully working prototype capable of ingesting and analyzing gigabyte-scale PCAPs
Developed a modular AI pipeline combining NLP, data parsing, and visualization
Created human-readable insights from raw network data, bridging the gap between engineering and trading teams
Demonstrated how AI can enhance transparency and compliance in financial markets

What we learned

How to design AI-assisted analytical tools for specialized domains like finance and networking
The importance of clean data pipelines and prompt engineering when working with domain-specific datasets
Effective use of Azure AI Foundry and Ollama to orchestrate local and cloud-based LLM reasoning
Real-world application of Streamlit and Python networking libraries for interactive analytics

What's next for Exchlytics AI

Integrate predictive modeling for latency and anomaly forecasting using time-series ML models
Expand to multi-exchange support with real-time PCAP streaming
Add compliance dashboards for SEBI and FINRA-style audit trails
Build an API layer to allow fintech platforms to plug into Exchlytics AI as an analytics microservice