Knowledge Assistant for Joint Utility: A Multi-Agent LLM-Driven Conversational System for Automated Task Execution
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The integration of Large Language Models (LLMs) into developer workflows has redefined automation, yet most systems remain limited to executing direct terminal commands. A critical limitation in existing tools is the lack of adaptive AI systems that cater to both technical and non-technical users. Knowledge Assistant for Joint Utility advances this capability by enabling not only command execution but also intelligent explanation, code comprehension, and handling of complex, context-dependent tasks through file analysis. The system functions as a fully embedded LLM within the terminal environment. This assistant introduces a modular, agent-based architecture that supports debugging, screen processing, code editing, web-assisted installations, and dynamic project integration through a FastAPI backend and Java-based graphical interface. The system’s core innovation is its autonomous self-customization engine, where the Builder Agent interprets natural language requests to generate, implement, and integrate new functionalities into the assistant’s own codebase, allowing the system to dynamically evolve. Specialized agents—including Scraper, LLM, Builder, and Assistant—coordinate to deliver iterative, context-aware development and intelligent task handling. Evaluation across 124 installations yielded a Task Success Rate of 88.75% using GPT-4o and a Completion Rate of 93.45%, outperforming other LLMs such as Gemini and DeepSeek. These results highlight the system’s potential as a multi-agent computational framework that integrates conversational interaction with autonomous software engineering.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it