.png&w=256&q=75)
🤓 Latest Submissions
.png&w=828&q=75)
TableKeeper
### Project Description: Tablekeeper **Tablekeeper** is a production-grade, autonomous restaurant reservation platform engineered to deliver seamless online table bookings while preventing race conditions and double-booking conflicts. Designed using a multi-agent **BAND Dark Factory** architecture, the platform separates responsibilities across four specialized seat mandates: Architectural Specification, Coder Implementation, Reviewer QA Testing, and Conductor Pipeline Orchestration. The backend is built with **FastAPI** and **SQLAlchemy**, utilizing SQLite WAL mode and atomic database transactions to guarantee strict concurrency defenses. Core features include real-time restaurant search, interactive time-slot availability checks, and complete reservation lifecycle management (creation, amendment, and cancellation). The application is paired with an aesthetic glassmorphism web dashboard for instant customer bookings and interactive OpenAPI documentation. Verified by an automated test harness with 100% test pass rates and high statement coverage, Tablekeeper represents a robust foundation for modern dining management systems.
6 Oct 2026
Recall — A Voice-First Personal Memory Agent
Recall is a voice-first personal memory agent designed to capture, organize, and retrieve knowledge from everyday conversations in real time. Built using Next.js 15, AssemblyAI Realtime Speech-to-Text WebSocket API, and Google Gemini LLMs, Recall acts as an intelligent second brain that actively listens to spoken audio, streams live transcripts, and automatically extracts structured facts, commitments, decisions, and named entities. Unlike traditional note-taking applications, Recall preserves explicit confidence states (CONFIRMED, LIKELY, UNCERTAIN) and dynamic temporal relationships. When new or updated information is discussed, Recall automatically reconciles updates and contradictions through relational linkage without destroying historical context, allowing users to explore an evolving timeline of knowledge over time. Equipped with a hybrid retrieval system and grounded conversational QA engine, Recall allows users to query their memories in natural language. It generates accurate, zero-hallucination answers backed by clickable source metadata and deep transcript line inspectors. Featuring a sleek, animated user interface, an interactive visual memory explorer filtered by people or projects, and a zero-dependency demo mode, Recall transforms spontaneous spoken dialog into an organized, persistent personal knowledge network.
30 Sep 2026
.png&w=828&q=75)
The Tribunal
**TRIBUNAL** is an autonomous AI hedge-fund-style trading system built on the core principle that no trade gets executed until it survives an adversarial trial. Traditional AI trading agents often suffer from confirmation bias—eagerly inventing reasons to buy whenever prompted. TRIBUNAL fixes this flaw by pitting specialized AI agents against each other in a formal courtroom setting. A **Prosecutor (Bull Agent)** builds the trade thesis, while a **Defense (Bear Agent)** actively searches for fatal flaws, valuation risks, and overbought indicators. **Quant, News, Macro, and Black Swan agents** supply data-driven evidence, technical calculations, and cross-examination. A **Judge Agent** synthesizes the debate to issue an `APPROVE`, `REJECT`, or `HOLD` verdict. Approved trades must then pass a non-LLM, hardcoded **deterministic Risk Gate** (enforcing position size, portfolio exposure, stop-loss, and buying power limits) before an order is submitted to **Alpaca Paper Trading**. Finally, a **Regret Engine** audits closed trades to score decision quality independently of raw P&L outcomes.
4 Sep 2026
