The problem
Signals were visible. Meaning was not.
The platform gave more than 500,000 monthly active users access to the kinds of trading signals professionals rely on, including candlestick patterns, price action indicators, and technical analysis metrics. But users without trading expertise could see those signals without easily understanding what they meant or what to examine next.
That interpretation gap created a shallow product experience. Users arrived, scanned the interface, and often left before working through the market context in depth. Average session duration stood at 3.2 minutes.
Why generic chat was not enough
- Generic LLM responses did not reliably follow the way traders evaluate real market signals.
- Standard chat interfaces did not combine indicators the way traders do.
- Off-the-shelf tools did not interpret candlestick patterns alongside price action, timeframe, and asset context.
The solution
Start with the trader’s decision process.
Techcreativ began with interviews with professional traders to understand the practical checks they make in live market conditions. Those interviews were converted into a pattern library that shaped both what the assistant analyzed and how it answered.
Market signals as callable tool agents
LangChain sits at the core of the assistant. Candlestick patterns and market signals were implemented as callable tool agents, allowing the model to route user questions to the appropriate analytical function during a conversation. Rather than generating market commentary from general model knowledge, the assistant runs pattern recognition on live data and ties responses to the current asset, timeframe, and signal context.
Conversational continuity
Entity extraction and memory track the assets, timeframes, and chart patterns a user has already discussed, allowing follow-up questions to build naturally on the previous interaction.
Production architecture
Each component had a clear role.
The outcome
More time spent analyzing, not just scanning.
Average session duration increased from 3.2 to 8.7 minutes.
On a platform with more than 500,000 monthly active users, the increase pointed to deeper product engagement. Users had more time to compare indicators, ask follow-up questions, and work through their interpretation of market conditions.