🤓 Latest Submissions
PROMETHEUS
PROMETHEUS is an autonomous competitive decision agent built for enterprises that need more than a report — they need a verdict. The system runs four agents in sequence, each with a distinct adversarial role: 1. SCOUT AGENT — deploys Gemini 2.5 Flash with live Google Search grounding to build a real-time evidence base: market data, competitor filings, analyst reports, news signals. Outputs a structured brief with citations. 2. CHALLENGER AGENT — reads Scout's findings and is explicitly instructed to find every reason the opportunity is overrated, the data is misleading, or the thesis is flawed. Real devil's advocacy — not polite hedging. 3. STRATEGIST AGENT — reads both Scout and Challenger. Produces an integrated strategic brief that identifies the core disagreements, weights them, and constructs actionable scenarios. Returns a DisagreementMatrix. 4. DECISION GATE — synthesises all three prior agents and produces the final verdict: GO, NO-GO, or MONITOR, with a calibrated confidence score (0–100%), key conditions, and explicit rationale. Three live scenarios are preloaded for instant demo: - OpenAI Market Entry → NO-GO at 65% confidence - Salesforce vs HubSpot Italy → MONITOR at 62% confidence - Tesla Competitive Threat Europe → GO at 88% confidence The UI populates each agent panel in sequence so judges can watch the reasoning unfold in real time — not just see the final answer. Tech stack: Gemini 2.5 Flash (all agents), Google Search grounding (Scout), FastAPI + Python backend deployed on Vultr Cloud Compute, Streamlit frontend on Streamlit Community Cloud, aiosqlite for analysis history. PROMETHEUS does what no existing AI tool does: it ends with a decision.
19 May 2026
Sentinel-Gate — Enterprise AI Security Gateway
Sentinel-Gate is a real-time security gateway for enterprise AI agents built for the Agent Security & AI Governance track powered by Veea. It sits between an enterprise application and the LLM, inspecting prompts before they reach the model and checking responses before they return to the user. The system uses Veea Lobster Trap as the inspection layer, Gemini AI for policy authoring and threat intelligence, Stream-lit for the operator interface, and SQLite for audit persistence. Security teams can test prompt injection, data exfiltration, HIPAA-style sensitive-data requests and custom policy violations from a live demo UI. Every verdict card shows the decision, risk score, detected intent and measured inspection latency in milliseconds. The Live Risk Monitor shows the latest risk score plus a sparkline of recent risk values, making attacker probing visible during the demo. Blocked prompts can open a Gemini threat intelligence report with attack type, severity, technique, remediation and compliance impact. Sentinel-Gate also includes compliance policy packs and plain-English custom policy creation. Every request is logged with timestamp, agent, prompt preview, intent, risk score and decision, with CSV export for review. The gateway is fail-closed: if inspection cannot certify a request, Sentinel-Gate blocks rather than passing unsafe traffic through.
19 May 2026
