ChaosPilot-Crash test for AI agents

A containerized OCR pipeline that reads US licence plates, Chinese plates and road signs from images, built on the mandated ROCm PyTorch base and run on AMD GPUs. Model: Qwen3-VL-8B-Instruct in bf16 with Hugging Face Transformers, greedy decoding. One prompt encodes the transcription rules: return only the registration number and drop state names and slogans around it, keep the Chinese province character and letter, copy characters exactly, and add no units or labels. Code post-processing joins multi-line text with single spaces and, for Chinese plates only, corrects O to 0 and I to 1 in the serial, which never uses those letters. Architecture: the harness starts a new process per image, and loading an 8B model takes longer than the 30-second limit. So the container's CMD starts a small server that loads the model once and warms up, and app.py is a thin client with a 25-second budget that always writes valid JSON, falling back to an empty answer if anything fails. Testing: beyond the 10 samples, I built two harder test sets: the samples under 10 degradations (blur, darkness, noise, skew, low resolution and combinations) and 46 synthetic plates and signs with unseen texts, also degraded. Results: 200/200 on the samples, 99/100 and 454/460 on the hard sets. Preprocessing with autocontrast was tried and dropped because it hurt accuracy. On an AMD Instinct MI300X: healthy after 23 s, 0.2-0.4 s per image, peak VRAM 18 GiB, image 45.7 GiB uncompressed. Built with AI assistance (Claude) for code review and test tooling.

Abrik-Ai

Ghostfox Zero is a two-match retrieval portfolio built entirely on AMD ROCm — one resident Qwen3-4B-Instruct server in bf16 on the AMD GPU, one BM25-first evidence pipeline, two containerized task front ends. MATCH 4 (Intelligent Web Information Retrieval): the agent navigates a saved copy of a website like a browser — every HTML page parsed, all navigation links followed, full BM25 index built over cleaned text, zero downloads at evaluation time. Given a natural-language query it extracts the exact value requested (name, date, price, version, phone number) with the source page as citation. MATCH 3 (Retrieval-Augmented Generation): the same fail-cited pipeline extended to mixed-document corpora — PDF (pypdf), DOCX and XLSX (stdlib zip+xml), CSV, source code, logs and config files all indexed with the same BM25 core, returning the short answer plus the exact files it came from, and nothing when the corpus does not contain the answer. Both containers implement the full league contract: phase 1 indexes the corpus and starts the resident model server within the startup budget, phase 2 answers each query as a separate process within the time limit, writing output JSON with answer, citations and confidence. Fail-closed by design: exact values only, citations always, empty answer instead of a guess — evidence-first where naive LLM agents hallucinate.

autokeren-solo

Maximum power, size flexible Is the $50 a credit balance inside Render $50 existing credit only Only when I open OpenCode Maximum power, size flexible Is the $50 a credit balance inside Render $50 existing credit only Only when I open OpenCodeMaximum power, size flexible Is the $50 a credit balance inside Render $50 existing credit only Only when I open OpenCodeMaximum power, size flexible Is the $50 a credit balance inside Render $50 existing credit only Only when I open OpenCodeMaximum power, size flexible Is the $50 a credit balance inside Render $50 existing credit only Only when I open OpenCodeMaximum power, size flexible Is the $50 a credit balance inside Render $50 existing credit only Only when I open OpenCodeMaximum power, size flexible Is the $50 a credit balance inside Render $50 existing credit only Only when I open OpenCodeMaximum power, size flexible Is the $50 a credit balance inside Render $50 existing credit only Only when I open OpenCodeMaximum power, size flexible Is the $50 a credit balance inside Render $50 existing

not sure

AetherTactics is an autonomous strategy AI agent engineered for the AMD AI League Match 6 challenge: "Can Your AI Master the Unknown?". The agent navigates an unfamiliar, procedural fog-of-war grid, evades environmental hazards, recovers mission-critical alien relics, and safely reaches extraction points while aggressively minimizing compute overhead and token consumption. Powered by Google Gemini (supporting Gemini 2.5 Flash, 3.1 Flash-Lite, and custom endpoints), AetherTactics utilizes an ultra-compact radar serialization protocol. Instead of verbose context passing, it translates 2D exploration states into dense, minified matrices, constraining LLM reasoning to strict micro-JSON directives under 60 output tokens. To prevent agent freezes during live competition, the platform incorporates an automated watchdog failover pipeline. If a model encounters 503 high-demand or rate-limit spikes, the system automatically executes exponential backoff and fails over to stable alternatives without interrupting autonomous navigation. Built-in live telemetry monitors prompt tokens, generation tokens, step latency, and turn efficiency to deliver state-of-the-art decision-making on minimal inference budgets.

Solo Builder XP

Most web agents summarize or hand back a list of links. Pinpoint returns the one value you asked for (a price, a date, a version, a phone number) together with the page and the exact quote it came from. If it cannot verify the value, it returns an empty answer instead of guessing. How it works: each question starts by writing a valid placeholder JSON, so a crash never breaks the output. The agent fetches the start page and up to 8 same-site links ranked by how well their link text matches the question. It switches to a headless Chromium browser only when a page renders with JavaScript. Pages are cleaned to text with table rows kept, and passages are ranked with BM25. A question classifier decides the expected answer type, and Qwen3-8B, served locally on an AMD Instinct GPU with ROCm, picks a candidate value from the top passages. The answer is accepted only if it appears verbatim in the cited passage and matches the expected type. Otherwise a rule-based extractor is tried, and then an empty answer. The model loads once at container start, and every question runs under a hard deadline. No external API keys are used: all inference runs on the AMD GPU. Results on our open 40-question benchmark across [X] real sites: [X]% exact match, 0 fabricated answers, 8/8 correct empty answers on unanswerable questions, median [X] s per question. The benchmark, eval script and Docker image are open source.

Pauprint