RB APPS

AI Agent / Working Prototype

Hermes AI
Trading Research Agent.

A personal AI market-analysis assistant built to analyze Indian markets, scan potential opportunities, apply predefined risk and quality filters, and deliver structured alerts through Telegram.

← Return to AI Agents

Overview

Hermes automates repetitive market research and delivers structured alerts through Telegram. It researches and filters opportunities while final decisions and real-money trades remain human-controlled.

Capabilities

What it does.

Scheduled market scansPre-market researchIntraday opportunity scanningEquity and index analysisOptions analysisRisk/reward filteringTelegram alertsNo-trade filtering

System Architecture

Research to alert.

  1. Market Information
  2. Market Research Tools
  3. Hermes AI Agent
  4. Custom Trading Skills
  5. Risk & Quality Filters
  6. Opportunity Scanner
  7. Telegram Bot
  8. User

Technology Stack

Persistent workflow.

  • Nous Hermes Agent
  • Cloud Reasoning Model
  • Telegram Bot
  • Scheduled Automation
  • Custom Agent Skills
  • Market Research Tools
  • LM Studio / Local Qwen Models

Risk Management

Selectivity before alerts.

  • Approximately 1% maximum risk per trade
  • Preferred minimum risk-to-reward: 1:2
  • No Trade when setup quality is insufficient
  • Human-controlled trade execution

Why not fully automated trading?

Research → Validation → Alert → Human Decision

Hermes does not automatically place real-money brokerage orders. Research and alerting remain separate from execution.

Engineering

Challenges and solutions.

Large context requirements

Solution: move reasoning to cloud infrastructure while Hermes remains the orchestration layer.

Hallucinated market data

Solution: analyze available data, report unavailable data, and never invent market data.

Excessive alerts

Solution: quality filters, risk/reward requirements, daily limits and silent execution when nothing qualifies.

Consistent analysis

Solution: reusable Hermes skills rather than a new prompt for every scan.

Architecture Evolution

Designed around constraints.

Initial

Hermes → LM Studio → Local Qwen Model

Final

Hermes → Cloud Reasoning Model

The local experiment could comfortably support around 8K context; larger contexts caused KV-cache memory allocation failures. Moving reasoning to the cloud improved reliability.

Project Gallery

Real prototype interfaces.

Screenshots show the project interface and prototype research outputs for technology and portfolio context, not financial-performance claims.

Hermes Agent terminal interfaceTelegram market brief from Hermes prototypeTelegram research output from Hermes prototypeTelegram validation output from Hermes prototype

Future Development

Planned, not implemented.

  • Paper Trading Engine
  • Backtesting
  • Trading Journal
  • Performance Analytics
  • Dashboard
  • Strategy Comparison
This project is presented for technology, research and portfolio purposes. It does not provide guaranteed returns or financial advice.