Comparative Analysis

Comparative Analysis: DeepDelve vs. LLM Wiki & Deep-Research-Agent

This document provides an in-depth comparative audit of the DeepDelve project against two state-of-the-art reference implementations:

1. nashsu/llm_wiki (Persistent knowledge-base/wiki paradigm)
2. CYC2002tommy/Deep-Research-Agent (Deep Science Writer academic pipeline)

It details the core conceptual differences, identifies functional gaps, and extracts concrete, high-impact features that can be implemented to transition DeepDelve from a search-harness into an industrial-grade research assistant.

---

1. Feature Matrix Comparison

Feature / Dimension	DeepDelve (Current)	LLM Wiki (nashsu)	Deep-Research-Agent (CYC2002tommy)
Primary Goal	General-purpose local deep web research	Incrementally built local wiki & knowledge base	Academic literature review & manuscript drafting
Output Format	Markdown (final_report.md)	Interlinked Obsidian-compatible Markdown pages	APA 7th edition .docx manuscript, Obsidian
Search Engine	DuckDuckGo Search (snippets + auto-fetch top-1)	Keyword + Vector hybrid search on local wiki	Academic databases (OpenAlex, Scopus, Semantic Scholar, Exa)
Knowledge State	Ephemeral (isolated session folders per run)	Persistent (living, incrementally updated wiki structure)	Ephemeral (focused study draft output)
Verification Gate	URL presence (fetched) + deterministic term overlap	Review queue for user-managed ingestion	Strict DOI lookup, Q1/Q2 journal check, peer-critique
Local Model Fit	Tuned for Mistral-Nemo (TUI & CLI modes)	API-integrated for Claude Code / desktop UI	Scoped for Hermes-3-Llama-3.1-8B (OpenClaw skill)

---

2. Deep-Dive: DeepDelve vs. llm_wiki

A. Core Paradigm Difference

• DeepDelve (Ephemeral RAG): Processes each query from scratch. When a user runs a query, it spins up sub-agents, searches the web, downloads files, writes a final report, and saves it in a timestamped folder (run_1783720884/). The knowledge cache is a simple {query -> text} map.
• LLM Wiki ( incremental Wiki): Treats the agent as a "librarian" that builds and maintains a living, persistent knowledge graph. Instead of writing standalone reports, the agent incrementally updates existing Markdown files, creates new interlinked pages, and updates indices and concept maps.

B. What DeepDelve Can Adopt from llm_wiki

1. Persistent Wiki Structure (Replacing Session Isolation)

Instead of archiving each run into a clean, isolated directory, we can support a Wiki Mode where the workspace is a persistent directory structure:

• wiki/ ( Obsidian-compatible folder)
• wiki/concepts/ (Topic-specific pages)
• wiki/sources/ (Markdown extractions of fetched sources)
• wiki/index.md (Living table of contents managed by the Planner)

2. Automatic Cross-Linking (Wikilinks)

When writing reports or concepts, the Planner should cross-reference other files in the workspace.

• Implementation: Modify the Planner's tools to read the list of existing page names, and instruct the Planner to format references as [[Page Name]]. Obsidian will automatically resolve these into a visual graph.

3. Living Wiki Schema (Control Files)

Introduce schema rules directly in the workspace (purpose.md and schema.md). The Planner reads these at startup to understand the style, naming conventions, and structure of the knowledge base, preventing it from producing unstructured or floating files.

---

3. Deep-Dive: DeepDelve vs. CYC2002tommy/Deep-Research-Agent

A. Core Paradigm Difference

• DeepDelve (General Web-Scraping): Relies on DuckDuckGo Search. Even the AcademicSearcher uses DDGS, meaning academic queries pull general web summaries, news, or blog posts instead of scholarly metadata.
• CYC2002tommy (Academic Rigor & Anti-Hallucination): Integrates directly with open scientific APIs (OpenAlex, Semantic Scholar, Scopus) to fetch peer-reviewed papers. It enforces strict DOI validation and screens journals by rank to ensure high-quality sources.

B. What DeepDelve Can Adopt from CYC2002tommy

1. OpenAlex / Semantic Scholar API Integration

Instead of forcing AcademicSearcher to use DuckDuckGo, we can add a dedicated academic retrieval tool.

• Implementation: Implement an academic_search tool that queries the OpenAlex API (free, no API key required for the polite pool). It allows searching papers by title, author, and concept, returning structured JSON containing DOI, journal index, and abstract.

2. Journal Quality Filtering (Impact Screen)

Many search results include articles from predatory or low-quality journals (e.g. MDPI, Frontiers, Q4 journals).

• Implementation: The academic_search tool can parse the journal ranking (Q1, Q2, Q3, Q4) or citation counts returned by OpenAlex and prioritize top-tier journals, preventing the agent from sourcing low-quality papers.

3. Strict DOI Verification Gate

Rather than just checking that a URL was fetched, verify the citation's Digital Object Identifier (DOI).

• Implementation: Create a lightweight verify_doi tool that queries api.crossref.org to confirm the DOI is valid and matches the paper metadata cited in the report.

4. Peer-Review Critique Phase (Nature/Science Level)

In DeepDelve's Planner workflow, Pass 2 is a "Global Critic" checking if claims exist in findings. We can elevate this to an academic peer-review simulation:

• Before compiling final_report.md, spin up a temporary sub-agent configured as a "Peer Reviewer" to critique findings.md. It evaluates:
  • Sample size limitations (e.g. "Source A claims X but only tested 12 subjects").
  • Overgeneralization of results.
  • Conflicts of interest (e.g. industry-funded papers).
  • Outdated findings (papers published >10 years ago).

---

4. Concrete Roadmap for Implementation

We can introduce these features in DeepDelve without breaking its core architecture by adding specialized tools and configuring new options in config.yaml:

Phase A: The Academic DB Toolkit (Academic Searcher Upgrades)

1. Tool academic_query_openalex(query: str, limit: int) [NEW]: Queries api.openalex.org for papers matching the terms. Returns:

   {
     "title": "...",
     "doi": "https://doi.org/... ",
     "authors": ["Author A", "Author B"],
     "journal": "Nature Communications",
     "quartile": "Q1",
     "abstract": "..."
   }

2. Tool verify_doi(doi: str) [NEW]: Performs a head request or lightweight CrossRef lookup to confirm validity.

Phase B: The Obsidian Wiki Module (Planner Upgrades)

1. Tool list_wiki_links() [NEW]: Returns list of all concept files in the workspace (e.g., ["Elasticsearch", "pgvector"]).
2. Planner Prompt Update: Update PLANNER_INSTRUCTIONS to dictate that when a concept from list_wiki_links is discussed, it MUST be wrapped in [[concept_name]] formatting.
3. Workspace Configuration: Add settings.workspace.mode: "wiki" (defaulting to "isolated"). When set to "wiki", session_isolation is bypassed, saving everything into a consolidated, interlinked workspace root.