🤓 Latest Submissions
Bee Sentinel-X
BEE SENTINEL-X is an autonomous, production-grade decision and paper-trading system designed for Kraken xStocks. Traditional algorithmic trading infrastructure often relies strictly on rigid mathematical formulas, completely missing breaking news catalysts. Conversely, human traders cannot monitor round-the-clock financial markets without emotional bias. Bee Sentinel-X bridges this gap by introducing a governed, multi-layered agentic loop that fuses live technical indicators with real-time unstructured financial sentiment analysis. Built on a highly modular Python 3.11 architecture, the system coordinates ingestion pipelines that calculate short-term moving averages (SMA-10/20) and RSI-14 indicators locally, while simultaneously scraping live financial headlines via Yahoo Finance RSS feeds. This compact, contextual context payload is mapped into a strict JSON schema and dispatched to the Gemini 2.5 Flash reasoning engine to generate programmatic BUY, SELL, or HOLD decisions alongside an explicit confidence score and structural rationale. Security and transparency are at the core of our framework. Before any trade hits the execution layer, the signal must clear the hardcoded "Sentinel Gate" policy engine. This risk manager enforces a strict 80% confidence threshold, filters overextended momentum states (RSI overbought/oversold protections), and prevents illegal short-selling by verifying simulated portfolio custody. Approved paper-trading commands are then asynchronously executed via the official Kraken CLI binary inside a containerized Docker environment. For enterprise-grade observability, every lifecycle step, model reasoning, and simulated execution is written into append-only JSONL audit logs, served directly on an interactive Gradio dashboard deployed on Hugging Face Spaces.
19 May 2026
Hive Industrial Sentinel
Energy and industrial operators need faster diagnostics across telemetry, manuals, and operating procedures. The same autonomy that makes agents valuable also creates risk: a prompt injection, hallucinated recommendation, or malicious operator command can push physical systems outside safe limits. Hive Industrial Sentinel demonstrates a safer pattern. It monitors transformer telemetry, detects thermal overload, retrieves guidance from engineering manuals, and recommends controlled mitigation. Before prompts or responses are trusted, a policy layer blocks requests to ignore alarms, disable safeguards, or force equipment above emergency limits.
19 May 2026
The Electrical Compliance Agent
The Electrical Compliance Agent is a hackathon MVP built for AI Agents and Agentic Workflows. The system receives a project description, normalizes it into auditable data, retrieves relevant evidence from a Supabase vector store of technical standards, generates a structured compliance report, and adds support suggestions for components when a finding clearly indicates a material or protection gap. The workflow consists of four specialized agents: Triage Agent: Receives and normalizes user input, extracting critical data such as power, voltage, and wire gauges. Research Agent (RAG): Performs semantic searches on a Supabase database to retrieve exact excerpts from the NBR 5410:2004 standard, eliminating hallucinations and ensuring a regulatory foundation. Audit Agent: Cross-references project data with retrieved evidence, generating findings classified by severity with direct references to standard clauses. Support Agent: Operates at the remediation layer, suggesting exact substitutes from a mock component catalog to fix detected non-compliance issues. This project demonstrates how AI agent orchestration can transform slow, manual processes into efficient, safe, and fully traceable engineering workflows, significantly reducing the risk of fatal errors in electrical installations.
10 May 2026
