How We Decode Startups

Data Collection

We monitor 50+ sources including funding announcements, company websites, technical docs, job postings, and press releases. Our crawlers extract structured and unstructured data that feeds into the analysis pipeline.

Architecture Detection

Each company goes through a multi-stage analysis using large language models. We extract stack signals, identify build patterns, assess GenAI depth, and surface implementation tradeoffs.

Pattern Taxonomy

Our pattern taxonomy represents recurring architectural decisions that signal how companies are building with AI. Each pattern includes:

• •Definition: What the pattern means
• •Implications: Considerations for builders and investors
• •Time horizon: Expected maturity timeline
• •Failure modes: What could derail the approach

Current patterns: Agentic Architectures, Vertical Data Moats, RAG, Micro-model Meshes, Continuous-learning Flywheels, Guardrail-as-LLM.

Atlas Brief Process

Each month, we synthesize individual company dossiers into an Atlas Brief:

1. 1Aggregate funding data and compute summary statistics
2. 2Identify pattern distribution and month-over-month changes
3. 3Generate thesis and implications for builders and investors using LLM synthesis
4. 4Editorial review for accuracy and signal quality
5. 5Publish by the 5th of each month

Evidence & Confidence

Every company dossier includes:

• •Evidence count showing how many source documents were analyzed
• •Freshness timestamp indicating when data was last updated
• •Source evidence quotes with citations
• •Confidence scores for pattern assignments