.png&w=256&q=75)
🤝 Top Collaborators
.png&w=640&q=75)
.png&w=640&q=75)
.png&w=640&q=75)
🤓 Latest Submissions
Redline
Redline is an audit layer for AssemblyAI’s Dictation API. The API returns two strings: text (verbatim transcript) and llm_response (LLM cleanup). Apps that paste only the rewrite can drop a negation, change a number, or collapse a spoken command and still get HTTP 200 with llm_error: null. Redline aligns those strings with a deterministic LCS checker (no second model), marks every span in the rewrite that does not trace back to the verbatim, and shows a verdict with findings. The demo workspace supports live record, an offline sample, and paste mode that runs entirely in the browser. Evidence is a 15-clip corpus (3 high, 1 medium, 11 clean) plus published false positives and known limits. Zero npm runtime dependencies: Node http server, vanilla HTML/CSS/JS, Web Audio AudioWorklet to WAV. Thesis: the cleanup can change what you said and this shows you when it does.
30 Sep 2026
Covenant: Proof-Carrying Trading Agent
Covenant is a proof-carrying options trading agent built for Alpaca Paper Trading. Instead of letting an AI model directly place orders, Covenant separates proposal, safety, authority, and execution into strict boundaries. A trader begins in Mandate Studio, where a plain-English risk mandate is compiled into a typed policy. The Alpha Engine then proposes only defined-risk SPY and QQQ vertical spreads. Before an order can move forward, the Safety Kernel evaluates it against eight deterministic invariants covering maximum loss, portfolio heat, daily halt conditions, market session state, liquidity, duplicate exposure, exit constraints, and verified account and market snapshots. A qualifying trade still cannot execute automatically. Covenant issues an Ed25519-signed TradePermit bound to the exact option legs, quantity, price band, policy version, and snapshot hashes. The permit expires after 60 seconds and has a single-use nonce, so replayed, expired, mutated, or oversized orders are rejected before they reach Alpaca. The model may propose, veto, shrink, and explain. It cannot activate policy, sign permits, widen risk, or submit orders by itself. Covenant makes paper-trading authority verifiable rather than assumed.
4 Sep 2026
Atlas AI
Atlas AI is an explainable artificial intelligence system designed for trading platforms operating in high-risk financial environments. Today, platforms face three major challenges: financial losses caused by fraud, poor user risk management, and increasing regulatory pressure requiring transparency in automated decisions. Existing solutions often rely on black-box models that lack trust and accountability. Atlas AI addresses this gap by combining real-time risk prediction with explainable AI. The system analyzes user behavior, trading patterns, and risk signals to detect anomalies, predict potential losses, and flag high-risk accounts. For every alert or decision, Atlas AI provides clear, human-readable explanations that can be understood by platform operators, users, and regulators. This approach improves trust, reduces fraud, strengthens compliance, and enables safer, more responsible platform growth. Atlas AI is scalable, API-driven, and designed to integrate seamlessly into platforms like Deriv.
7 Feb 2026
AI Chief of staff
AI Chief of Staff is an enterprise-grade, fully autonomous agentic AI platform built on IBM watsonx Orchestrate. Designed to function like a virtual Chief Operating Officer, it intelligently analyzes organizational KPIs, identifies operational bottlenecks, generates weekly strategies, and orchestrates tasks across teams — transforming executive operations into a self-driving system. With a seven-agent multi-layer architecture, the system automates every stage of operational management: KPI Data Ingestion: Reads and validates business metrics from CSV or Google Sheets. KPI Analysis: Computes trends, anomalies, and week-over-week performance changes. Root Cause Diagnosis: Determines the most likely reasons KPIs underperform using pattern analysis. Strategy Planning: Generates prioritized initiatives with owners, acceptance criteria, risks, and effort estimations. Task Orchestration: Creates Jira/Trello tasks, assigns owners, sets due dates, and updates Notion docs. Meeting Scheduling: Automatically coordinates team calendars and creates meeting invites with agendas. Risk Prediction: Calculates SLA breach probability and recommends mitigations. This system reduces executive planning time by 70%, automates 90% of routine execution tasks, and identifies risks days before they escalate, offering measurable ROI for any organization. Built for the Agentic AI Hackathon with IBM watsonx Orchestrate, the platform demonstrates how workflow automation, LLM reasoning, and tool integrations (Jira, Slack, Google Calendar, Notion, Sheets) can collaborate to create a next-generation digital executive. The result: A truly autonomous COO that turns organizational data into decisions, strategies, tasks, and actions — all in less than 90 seconds.
23 Nov 2025
