ThinkStorm · Devlog
Architectural Foundation: Frictionless Intake, Multi-Stage Pipelines, and Untrusted URL Boundaries
Established the core ThinkStorm architecture, featuring frictionless anonymous idea capture, VirusTotal URL scanning, bounded multi-stage research pipelines, prompt catalog versioning, and token usage accounting.
Architectural Foundation: Frictionless Intake & Multi-Stage Pipelines
The core design objective of ThinkStorm is solving the cognitive barrier of initial idea capture. Most project management and idea tracking platforms burden users with mandatory fields, rigid taxonomies, and categorization upfront. ThinkStorm flips this model: users throw free-form thoughts into an anonymous intake box with zero friction, while autonomous background processors handle structuring, research, and technical evaluation.
What Was Implemented
1. Bounded Intake Engine & Safety Boundaries
- Implemented single-textarea intake accepting raw text up to 10,000 characters.
- Built an automatic URL extraction engine that isolates embedded links from untrusted submission text.
- Integrated asynchronous VirusTotal scanning that evaluates link reputation, domain risk profiles, and malicious redirection vectors. Submissions containing flagged URLs are automatically routed to a secure quarantine state for administrator review before any model processing runs.
2. Multi-Stage Pipeline Orchestration
- Processor 1 (Safety Evaluator): Validates link safety and determines automated processing eligibility.
- Processor 2 (Semantic Extraction): Extracts concise project titles, executive summaries, categories, and tags while preserving the raw submission text immutably.
- Processor 3 (Duplicate & Relationship Detector): Computes semantic overlaps and links related ideas in SQLite.
- Processor 4 (Prior Art & Landscape Search): Interfaces with SearXNG and Perplexica to discover live competitors and open-source equivalents.
- Processor 5 (Technical Feasibility & Critique): Generates rigorous feasibility scoring, architectural constraints, and risk matrices.
3. Prompt Catalog & Provenance Telemetry
- Modeled prompt definitions with semantic versioning (
prompt_definitionsandprompt_versions). - Tracked complete execution provenance (
processor_runs) capturing model policies, resolved backend providers, input/output token counts, durations, and output artifacts.