Cascade Trend Analysis

An intelligent system for analyzing technology trends and forecasting their cascading impact. Automatically collects signals from multiple sources, groups them into trends, evaluates them using a three-dimensional model (significance, momentum, confidence), and generates analytical reports with actionable recommendations.

---

What the system does

• Signal collection from 5 sources: Hacker News, GitHub, arXiv, Semantic Scholar, SearXNG (246+ search engines)
• Source-aware scoring — separate formulas for each source type with web citation enrichment
• LLM classification — grouping signals into trends with automatic category detection
• 3D trend profiling — evaluation across three axes: Significance / Momentum / Confidence
• Recommendations — automatic calculation: ACT_NOW, MONITOR, RISKY_HYPE, EVERGREEN, IGNORE
• Analytical reports — cascading impact graph, impact zones, validated sources
• Role-based recommendations — specific actions for CTO, Developer, PM, Investor per impact zone
• Multilingual support — context-aware batch translation (EN, RU) with automatic language detection
• External API — REST API v1 with Personal Access Token authentication for external agent integration

Key Features

Scoring and Recommendations

Each trend is evaluated using a three-dimensional model:

Dimension	What it evaluates	Weight
Significance	Evidence strength, source diversity, signal density	40%
Momentum	Signal arrival rate, score trajectory, freshness	35%
Confidence	Source coverage, sample size, data consistency	25%

Based on axis combinations, the system automatically determines a recommendation and lifecycle phase (emerging → mature → fading).

Cascading Analysis

LLM agent (LangGraph + Claude) generates:

• Full analytical report with trend overview
• Cascading impact graph — how the trend propagates to adjacent areas
• Impact zones with influence rating (1-10)
• Role-based recommendations — specific actions for 4 roles (CTO, Developer, PM, Investor)
• Sources validated via web citations

Zone ID Filters (v0.15)

• Impact zone filtering (integer ID) on /trends, /signals, /pulse — language-independent
• Pulse heatmap — activity visualization by zones and categories
• Zone filter in reports with Popover search

Translation

Context-aware batch translation using LLM:

• Titles translated together with descriptions and categories for accuracy
• Automatic source language detection (Cyrillic / Latin)
• 2 LLM calls per analysis instead of 30-80 (batched short strings with deduplication)

External API

REST API v1 for external agents and integrations:

• GET /api/v1/trends — paginated list with sorting
• GET /api/v1/trends/top — Top 5 by 3 criteria (new, fast_growing, highest_scored)
• GET /api/v1/trends/{id} — trend details with signals and latest analysis
• GET /api/v1/signals — signal list with filters
• Authentication: Personal Access Tokens (PAT)

Architecture

Frontend (React + TanStack Router + shadcn/ui)
    ↕ REST API
FastAPI + SQLite
    ├── Crawler → 5 source adapters → signals table
    ├── Scoring → source-aware + web citation enrichment
    ├── Classifier → LLM trend grouping + categorization → trends table
    ├── Scorer → 3D model (Significance/Momentum/Confidence) → recommendations
    ├── Analyzer → LangGraph + Claude → reports + impact zones
    ├── Recommendations → role-based actions (CTO/Dev/PM/Investor)
    ├── Translator → context-aware batch translation → translations table
    └── External API v1 → PAT auth + rate limiting
External:
    ├── SearXNG (self-hosted meta-search, Docker)
    ├── Claude API (Anthropic) — analysis + classification + translation
    └── FAISS — vector similarity

Tech Stack

Component	Technology
Backend	Python 3.12+, FastAPI, uvicorn
Database	SQLite (single file)
Frontend	React, TypeScript, TanStack Router, Zustand, shadcn/ui
LLM	Anthropic Claude (via LangGraph)
Search	SearXNG (self-hosted, Docker)
Vector Store	FAISS
CI/CD	GitLab CI, PM2

---

Version: 0.15.14 | March 2026

v0.15.14 — March 2026

Zone ID filters, pulse heatmap overhaul, API modularization

New Features

• Zone ID filter — filter by impact zones on trends, signals, and pulse pages. Language-independent.
• Reports zone filter — reports page supports impact zone filtering with search.
• Automatic zone resolution — impact zones are automatically linked to records after analysis completion.
• API modularization — API routes split into 6 specialized modules.
• Pulse heatmap — heat map displays activity across impact zones and categories.

Analysis Settings

• Cascading analysis depth: 2 levels
• Impact threshold: 0.3
• Improved zone matching: cross-category embedding search

Fixes

• Latest analysis reference updated on completion
• Backward compatibility for web citation reading

---

v0.14.0 — March 2026

Server-side report pagination, Lab: saved products

New Features

• Reports: server-side pagination and search — full-text search across analyses (EN + translations), sorting by date and score, configurable page size.
• Lab: saved products — save products to favorites with a dedicated saved items page.
• Trend name translation in Lab — multilingual support in the Lab module.

Fixes

• Fixed report sorting by date
• Fixed React hooks order
• Correct display of product cards without images

---

v0.13.0 — March 2026

Signal, role, and performance fixes

• Shared report — accessible via link without authentication
• Admin: data source management
• SearXNG remap optimization