From last few days i was experimenting / evaluating the trading strategies using AI.
I have tested the major LLMs like Claude, Qwen, Gemini, GPT, and DeepSeek. My conclusion? None of them can replace lived market experience.
The Core Problem: Context Window
This is where LLMs start breaking down. I experimented with Claude Skills, custom agents, detailed rule sets, and command structures and results were okay-ish at best, not reliable enough where real money is on the line.
A Real Example (Data from Dhan APIs)
I ran several stocks through my strategy, fed 6 months of data into multiple AI models, and asked them to rank the top 3 picks for the day.
The AI output:
APEX → Score 4.8
Privi Speciality Chemicals → Score 4.0
Global Education–> Score 3.8
Result was , Two of the three were down ~2%. Only Privi moved up, a modest ~0.60%.
Meanwhile, I independently identified MTAR Tech using my own analysis. It closed up ~8.33%.
My Takeaway is that AI is not useless for trading, but it is a very sophisticated IF-ELSE engine. It takes rules literally, struggles with nuance, and breaks down at the edges where markets actually live.
For now, my approach is clear: Human in the Loop.
AI handles the calculations, screening, and data crunching. I make the final call.
One more thing worth noting that most of the new AI products flooding the market are essentially wrappers around the same foundational models. Do not expect dramatically different results across them.
Would love to hear your thoughts and experiences. Are you using AI in your trading workflow? What is working for you? Also Interested to know if @Dhan team is working on something for this?