Live AI product · 2025-present
Apex
Market intelligence without the analysis overload.
- Context
- AI Product Manager · Product Designer · Full-stack builder
- 01
- ~95% lower LLM spend
- 02
- 4 asset classes
- 03
- ~30 sec structured reads
- 04
- 500+ crypto pairs
Problem
Traders face too many indicators, conflicting analysts, and emotionally charged decisions. Generic AI adds prose, but it does not add a reliable analytical foundation.
Approach
I translated market logic developed over several years from Pine Script into Python. Apex processes live data first, then gives a smaller LLM structured evidence to explain in clear human language.
Outcome
A live credit-based product covering four asset classes, with decisive output in about 30 seconds and approximately 95% lower LLM spend.
Key product decisions
Market intelligence without the analysis overload.
- 01
Algorithms before language
Deterministic preprocessing does the analytical work; the LLM explains structured evidence instead of guessing from raw data.
- 02
High-value inference only
Always-on content uses rule-based NLG at near-zero model cost. Paid inference is reserved for deeper, on-demand analysis.
- 03
Clarity over terminal complexity
The product gives a verdict, rationale, levels, trade plan, and counter-case instead of another wall of indicators.
- 04
Support emotional moments
A dedicated Panic Button helps users step back and reassess a position when urgency can override judgment.
My scope
- Product thesis, roadmap, pricing, and launch
- End-to-end UX and interface design
- Prompt architecture, guardrails, and output format
- Python/FastAPI data and analysis services
- React/TypeScript frontend and production deployment
Product evidence
