Every AI project starts with a document. A requirements spec outlining what you want to build. A design document mapping your architecture. A deployment plan specifying your infrastructure. These documents shape billions of dollars in technology decisions, yet most teams write them in isolation—without access to the latest benchmarks, without knowledge of compatibility constraints, without awareness of emerging alternatives.
The gap between what AI teams know and what they need to know is widening. New models launch weekly. Benchmark leaderboards shift monthly. Compatibility matrices grow more complex. By the time you finish writing a requirements document, the assumptions behind it may already be outdated.
Neo document view showing a requirements document with improvement suggestions and agent workflow
Neo bridges that gap. Built on NeoSignal's knowledge graph of AI infrastructure components, benchmarks, and market signals, Neo deploys four specialized agents that transform how you create and refine AI solution documents. Upload a requirements spec, and the Reviewer agent scores it across four dimensions. The Advisor agent suggests improvements backed by live data. The Generator agent fills missing sections with citations. The Automator agent discovers opportunities to pre-populate NeoSignal's calculator tools with values extracted from your document.
The result: documents that reflect current reality rather than stale assumptions. Decisions grounded in data rather than intuition.
The Four Agents
Neo's architecture follows Anthropic's principles for building effective agents: single-purpose systems with clear responsibilities, explicit tool access, and structured outputs for downstream processing. Each agent specializes in one aspect of document intelligence.
| Agent | Role | Key Capabilities |
|---|---|---|
| Reviewer | Find Issues | 4-dimension scoring (completeness, accuracy, best practices, compatibility), severity-ranked issues |
| Advisor | Suggest Fixes | Prioritized improvements with NeoSignal citations, alternative component recommendations |
| Generator | Create Content | Missing sections, task lists, component recommendations with live data references |
| Automator | Configure Tools | Discovers calculator opportunities, pre-populates inputs from document context |
Save & organize insights
Save articles and excerpts to your personal library
Reviewer: Document Analysis
The Reviewer agent analyzes your document and produces a structured assessment. It scores four dimensions on a 0-100 scale: completeness (are all necessary sections present?), accuracy (do claims match current data?), best practice alignment (does the approach follow industry standards?), and compatibility (do the specified components work well together?).
Beyond scores, the Reviewer identifies specific issues. Each issue includes a type (gap, inaccuracy, best practice violation, or compatibility concern), severity level (high, medium, low), location in the document, description of the problem, and evidence from NeoSignal's knowledge base. A high-severity accuracy issue might note that a document claims GPT-4 leads code generation benchmarks when current data shows Claude Sonnet 4 has overtaken it on SWE-Bench.
The Reviewer uses four tools: componentLookup to search NeoSignal's database of models, accelerators, cloud providers, frameworks, and agents; benchmarkQuery to retrieve Epoch AI performance data; compatibilityCheck to verify component pairings; and ragRetrieval to search curated knowledge for best practices. These tools execute in parallel when queries are independent, minimizing analysis time.
Advisor: Improvement Suggestions
The Advisor agent transforms identified issues into actionable improvements. It prioritizes suggestions by impact—high-severity issues affecting core architecture decisions come before minor formatting concerns. Each improvement specifies what to change, why to change it, and what data supports the recommendation.
Neo improvements panel showing accepted suggestions with confidence scores and citations
Improvements come with confidence scores (0-100) reflecting how strongly the evidence supports the suggestion. A recommendation to switch inference providers based on benchmark data might score 92% confidence. A suggestion to add a missing security section based on best practices might score 78%. You can batch-apply improvements or review them individually, accepting some and rejecting others.
The citation system grounds every suggestion. Sources include NeoSignal components (with links to component detail pages), market signals (with confidence scores and data points), knowledge articles (from authoritative sources like the HuggingFace Ultrascaling Playbook), and benchmark data (from Epoch AI evaluations). When the Advisor suggests using vLLM over TensorRT-LLM for your use case, it cites the specific compatibility scores and latency benchmarks supporting that recommendation.
Generator: Content Creation
The Generator agent creates new document content based on your instructions. Request a missing "Security Requirements" section, and it produces structured content drawing from NeoSignal's knowledge base. Ask for a technology comparison table, and it queries component data to populate current scores and metrics.
Generated content matches your document's existing style. If your document uses numbered lists for requirements, generated sections use numbered lists. If your document follows a specific heading hierarchy, generated content respects that structure.
Like the Advisor, the Generator cites its sources. A generated component recommendation section might reference the Memory Calculator for GPU sizing, the TCO Calculator for cost projections, and specific NeoSignal components for the recommended stack. You can trace every claim back to its source.
Automator: Tool Discovery
The Automator agent bridges documents and tools. It analyzes your document, identifies mentions of AI infrastructure concepts (models, GPUs, batch sizes, token counts), and matches them to NeoSignal's calculator tools.
Neo timeline showing the Automator agent discovering 8 automation opportunities
For a document describing fine-tuning Llama 3.1 70B on 8x A100 GPUs with batch size 4, the Automator might identify opportunities for the Memory Calculator (pre-populated with model parameters and batch size), the Parallelism Advisor (pre-populated with cluster configuration), and the TCO Calculator (pre-populated with infrastructure requirements).
Each opportunity includes a confidence score, the extracted input values, and an explanation of what benefit the tool provides. Switch to the Automate tab to select opportunities and generate pre-configured tool artifacts. One click creates a Memory Calculator artifact with your document's specifications already filled in—no manual data entry required.
Knowledge Graph Integration
Neo's intelligence comes from NeoSignal's comprehensive knowledge graph. The platform maintains data on AI components across five categories: models (from Claude Sonnet to Llama to Gemini), accelerators (from H100 to TPU v5 to Trainium), cloud providers (from AWS to CoreWeave to Lambda Labs), frameworks (from vLLM to LangChain to PyTorch), and agents (from Claude Code to Cursor to Devin).
Each component includes a NeoSignal score (0-100), metrics specific to its category, compatibility mappings to other components, and trend indicators showing market momentum. When Neo analyzes a document mentioning "Claude 3.5 Sonnet," it looks up current benchmark scores, checks compatibility with specified inference frameworks, and identifies any recent signals affecting that model's competitive position.
Beyond components, the knowledge graph includes benchmark data from Epoch AI's evaluations (covering reasoning, mathematics, coding, agents, and specialized domains), market signals (leader changes, emerging players, trend shifts, compatibility alerts), and curated knowledge from authoritative sources like LMArena, SemiAnalysis, and official documentation.
The agents query this knowledge in real-time. When you upload a document today, Neo analyzes it against today's data—not a snapshot from weeks ago. If a new model launched yesterday and already appears in NeoSignal's database, Neo can reference it in suggestions.
What Makes Neo Different
Every Suggestion Includes Citations
Generic AI assistants generate plausible-sounding advice without sources. Neo grounds every recommendation in verifiable data. The citation system uses structured references that render as clickable links: component references link to NeoSignal component pages, signal references link to market intelligence entries, knowledge references link to source articles. You can verify any claim with one click.
Severity Scoring Prioritizes Your Attention
Not all issues deserve equal attention. A missing requirements section blocking stakeholder alignment matters more than a suboptimal phrasing choice. Neo ranks issues by severity and surfaces high-impact problems first. You spend time on decisions that matter rather than getting lost in minor suggestions.
Multi-Step Reasoning Verifies Claims
Neo agents don't just pattern-match against your document. They make tool calls to verify claims against live data. When your document states a model achieves certain benchmark scores, the Reviewer queries the benchmark database to confirm accuracy. When your document specifies a technology stack, the compatibility checker validates that components work well together. This multi-step reasoning catches issues that surface-level analysis would miss.
Batch Apply Gives You Control
You decide which improvements to accept. The Improve view presents all suggestions with their rationale and confidence scores. Accept improvements individually, batch-accept by type, or batch-accept all pending suggestions with one click. Rejected improvements disappear from the queue. The Diff view shows exactly what changed, with additions highlighted in green and deletions in red.
Neo diff view showing document changes with additions and deletions highlighted
Workflow Views
Neo provides six views for working with your document:
Doc shows the rendered document with improvement suggestions inline. Colored pills indicate suggested additions, modifications, and deletions. The sidebar tracks pending suggestions with quick accept/reject actions.
Timeline shows the agent workflow as a chronological sequence. Watch agents queue, run, and complete. Expand tool call details to see exactly what each agent queried and found. The timeline provides transparency into Neo's reasoning process.
Improve lists all improvement suggestions with full context. Filter by type (addition, modification, deletion) or status (pending, accepted, rejected). Batch actions let you process multiple suggestions at once.
Diff shows a unified diff of all accepted changes. Green highlights mark additions; red marks deletions. Review the cumulative impact of improvements before finalizing.
Graph visualizes the workflow as a node graph with swim lanes for user actions, agent tasks, outputs, and artifacts. Track how your document evolved through the agent pipeline.
Neo graph view showing workflow visualization with user, agent, and output lanes
Automate shows discovered tool opportunities from the Automator agent. Select opportunities to generate pre-configured calculator artifacts. View created automations with links to the corresponding tools.
Credit System
Neo uses NeoSignal's credit system for document processing:
| Workflow | Credits |
|---|---|
| Quick Review | 1 credit |
| Full Review (Reviewer + Advisor) | 2 credits |
| Generation | 3 credits |
| Automation Discovery | 1 credit |
The agent bar at the top of the Neo interface shows estimated processing time and word count. Larger documents take longer to analyze but don't cost more credits—the credit cost is per workflow, not per word.
Getting Started
Navigate to /neo to access Neo. The interface accepts document uploads via drag-and-drop or file picker. Supported formats include markdown (.md), plain text (.txt), and common document formats.
Once uploaded, Neo parses your document, identifies its type (requirements, design, spec, plan, or other), and extracts metadata including word count, headings, and any mentioned components. The agent workflow starts with the Reviewer—click through to Advisor, Generator, and Automator as needed for your use case.
Processed documents save as artifacts in your NeoSignal account. Return to previous documents from the Saved tab. Share artifact links with team members who have NeoSignal accounts.
The Document-First Future
AI infrastructure decisions increasingly flow from documents. Procurement specs determine vendor selection. Architecture documents shape deployment patterns. Requirements specs drive implementation choices. The quality of these documents directly impacts the quality of the resulting systems.
Neo represents a new approach to document intelligence: specialized agents with deep domain knowledge, working together to transform how teams create and refine AI solution documents. Not generic writing assistance, but infrastructure-aware analysis grounded in live data from NeoSignal's knowledge graph.
Upload a document. Run the agents. See what Neo finds. The gap between what you know and what you need to know just got smaller.
Related Features:
- NeoSignal Tools Hub - Calculator tools that Neo's Automator can pre-populate
- NeoSignal AI Chat - Context-aware assistance with citations
- NeoSignal Agents - Compare AI agent systems and capabilities