scholaredge system instructions

    1

    On-device academic research copilot for Snapdragon X Elite featuring hybrid RAG, strict grounding, and study tools.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    scholaredge system instructions

    Example session with this skill installed

    Compare the methodologies and quantitative results for "Attention is All You Need" and "BERT: Pre-training of Deep Bidirectional Transformers".

    • Read your context and instructions
    • Compiled the scholaredge system instructions

    I have analyzed the provided documents for "Attention is All You Need" and "BERT: Pre-training of Deep Bidirectional Transformers". The following comparison focuses on their architectural methodologies and primary benchmark performance.

    Research Paper Comparison Matrix

    DimensionAttention is All You Need (Transformer)BERT (Bidirectional Transformers)
    Core ObjectiveEliminate recurrence/convolution in favor of self-attention [Doc: Transformer, Page: 1, Chunk: 02]Pre-train deep bidirectional representations from unlabeled text [Doc: BERT, Page: 1, Chunk: 01]
    ArchitectureStacked self-attention and point-wise, fully connected layers for encoder/decoder [Doc: Transformer, Page: 3, Chunk: 08]Multi-layer bidirectional Transformer encoder based on the original Transformer [Doc: BERT, Page: 3, Chunk: 12]
    MethodologyScaled Dot-Product Attention; Multi-Head Attention; Positional Encoding [Doc: Transformer, Page: 4, Chunk: 15]Masked LM (MLM) and Next Sentence Prediction (NSP) tasks [Doc: BERT, Page: 4, Chunk: 18]
    Key Results28.4 BLEU on WMT 2014 English-to-German task [Doc: Transformer, Page: 7, Chunk: 22]80.5% GLUE score; SOTA on 11 NLP tasks [Doc: BERT, Page: 7, Chunk: 25]
    LimitationsScaling to very long sequences due to quadratic complexity [Doc: Transformer, Page: 8, Chunk: 30]High computational cost during pre-training; NSP task utility debated [Doc: BERT, Page: 9, Chunk: 34]

    Next steps

    1. Generate active learning flashcards for the BERT "Masked LM" mechanism.
    2. Extract the ablation study tables from both papers for specific hyperparameter comparison.
    3. Summarize the "Positional Encoding" section from the Transformer paper.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Academic research and study workflows often suffer from hallucination and privacy risks when processed through standard LLMs. This skill transforms your agent into a grounded research copilot optimized for the Snapdragon X Elite platform. It enforces strict citation requirements and local-first execution to ensure your private data stays on-device while providing high-precision analysis.

    What it does

    • Hybrid Retrieval combines BM25 lexical search with dense vector embeddings to find exact technical terms and semantic concepts.
    • Strict Grounding prevents hallucinations by restricting responses to provided document chunks with mandatory citation anchors.
    • Comparative Analysis generates structured matrices to evaluate methodologies, architectures, and results across multiple papers.
    • Active Learning Generation produces flashcards, self-assessment quizzes, and study outlines directly from ingested PDFs.
    • Multimodal Correlation links figures, tables, and charts to their textual descriptions for accurate visual comprehension.

    How it works

    1. Ingestion: The agent parses PDFs into structural blocks, maintaining section boundaries and metadata.
    2. Indexing: Documents are processed using sliding-window semantic chunking and indexed for hybrid fusion search.
    3. Querying: The agent executes RRF (Reciprocal Rank Fusion) to retrieve the most relevant context windows for your prompt.
    4. Synthesis: Responses are generated using only the retrieved context, complete with page and chunk-level citations.

    Frameworks & tools

    Built for Snapdragon X Elite platforms using Qualcomm QNN (Hexagon NPU) and ONNX Runtime. It utilizes all-MiniLM-L6-v2 for embeddings and integrates with SQLite for local vector storage.

    Why this beats prompting it yourself

    Generic prompts often lead to "lazy" RAG that misses specific mathematical notations or hallucinates missing details. This skill implements a sophisticated hybrid retrieval pipeline and rank fusion logic that a standard prompt cannot replicate reliably.

    Use cases

    • Literature Review to compare architectures and quantitative results across a library of research papers.
    • Exam Preparation by generating targeted flashcards and quizzes from course materials and textbooks.
    • Technical Auditing to verify architectural claims against specific figures and tables in technical documentation.

    Known limitations

    Optimized for Snapdragon X Elite hardware. Local embedding generation requires ONNX Runtime support. Performance may vary on non-NPU hardware.

    Requires NPU or ONNX Runtime for local embedding. Retrieval performance depends on PDF structural clarity. Hallucination prevention is limited to the provided document context.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 24 days ago

    • Free with an Agensi account

    Permissions

    Network Access
    Listed24 days ago

    What's inside

    Frequently Asked Questions

    ScholarEdge utilizes the Hexagon NPU on Snapdragon X Elite chips to perform local vector search and RAG synthesis, ensuring your private research documents never leave your device.

    The skill is optimized for any AI agent framework that supports ONNX Runtime and local file system access, specifically targeting Windows on Snapdragon environments.

    Unlike standard prompting, this skill uses a specialized hybrid retrieval pipeline (BM25 + Dense Vector) and Reciprocal Rank Fusion to ensure technical terms and mathematical formulas are accurately captured.

    You will receive the system instruction set, the pre-configured indexing logic for local vector storage, and the integration schema for Snapdragon's QNN execution.

    Yes, the skill includes a multimodal correlation module that links textual descriptions to tables and figures within the PDF structure for accurate technical auditing.

    While the skill can run on standard CPUs, it is specifically tuned for the Snapdragon X Elite NPU; users on non-NPU hardware may experience significantly slower indexing and response times.

    New in Research & Analysis