Abstractly

A research radar for tracking scientific trends and translating emerging techniques into practical applications. Aggregates newly published research from open-access sources across 15 disciplines and generates AI-assisted digests with multiple perspectives.

---

Features

Current

• 15 research disciplines — mathematics, CS, neuroscience, economics, law, arts, and more
• 51 configurable sources — arXiv (Atom), bioRxiv/medRxiv (JSON), OSF Preprints (JSON:API), Europe PMC (JSON)
• Browse and enable/disable disciplines, configure sources per discipline
• Preview raw feed entries from any source
• AI-summarized digests grouped by discipline and source with three perspectives per paper:
  • ELI5 — plain-language explanation
  • Solo SWE — developer opportunity framing
  • Investor — high-risk opportunity thesis
• Progressive streaming — digest sections render as each discipline completes (Livewire 3 $this->stream())
• Paper deduplication — cross-listed papers collapsed within each discipline with "Also in" badges
• Source caching — configurable TTL to avoid re-fetching (bypass via "Skip cache" checkbox)
• AI summary caching — cached per paper URL to avoid re-summarization on repeat generations
• Rate limiting — configurable inter-batch delay with 429 retry and exponential backoff
• JSON export — download digest as self-describing JSON with metadata envelope (timestamp, disciplines, sources, format version)
• Saved papers — bookmark papers from the digest, view on dedicated /saved page, persisted to local JSON file across sessions
• Visual progress bar — animated bar with source counter during digest generation
• Error visibility — amber warning banner listing failed sources, with specific messaging for API rate limits
• Session lifetime warnings — help text on picker pages indicating session-based persistence
• Multi-provider AI support (Gemini, OpenAI, Ollama) with graceful degradation on failure

Planned

• Research radar view with trend tracking
• Personal trend tracking
• Ranking by novelty or citations
• Cross-discipline clustering
• Integration with The Shelf for research-to-project pipeline

---

Architecture

The digest generation pipeline is a three-pass loop per discipline:

1. Fetch — SourcePreviewer pulls papers from 5 different API formats (Atom, JSON, JSON:API, REST, RSS) with per-source caching
2. Deduplicate — PaperDeduplicator collapses cross-listed papers across sources within each discipline, annotating first occurrences with "Also in" references
3. Summarize — AiSummarizer batches papers to the configured AI provider (Gemini/OpenAI/Ollama) with per-paper summary caching

Results stream to the browser progressively via Livewire 3's $this->stream() as each discipline completes — no queue infrastructure needed.

No database by design. This is a local-only, single-user research tool. File-based sessions, file-based cache, and a flat JSON file for saved papers are sufficient and keep the setup trivial. The deliberate trade: no multi-user support in exchange for zero database configuration and instant portability.

---

Tech Stack

• Backend: Laravel 12 (PHP 8.2+)
• Frontend: Livewire 3 (full-page components), Tailwind CSS v4, Alpine.js, Vite 7
• Storage: File-based sessions and cache (no database)
• AI Providers: Google Gemini (default), OpenAI, Ollama
• Data Sources: arXiv (Atom), bioRxiv/medRxiv (JSON), OSF Preprints (JSON:API), Europe PMC (JSON)

---

Quickstart

Docker (recommended)

git clone https://github.com/akuligowski9/abstractly.git
cd abstractly
cp .env.example .env    # set GOOGLE_API_KEY (or other AI provider keys)
docker compose up --build
# → http://localhost:8000

Saved papers, cache, and sessions persist across restarts via a named volume. To reset all data: docker compose down -v.

Local

# Clone
git clone https://github.com/akuligowski9/abstractly.git
cd abstractly

# Install dependencies and set up
composer setup

# Or manually:
composer install
cp .env.example .env
php artisan key:generate
npm install
npm run build

# Start development servers
composer dev

This runs the Laravel server, log tail, and Vite dev server concurrently.

---

Environment Variables

Copy .env.example to .env and configure:

Variable	Required	Description
GOOGLE_API_KEY	Yes (if using Gemini)	Google AI API key for Gemini
DIGEST_AI_PROVIDER	No	gemini (default), openai, or ollama
DIGEST_AI_MODEL	No	Gemini model (default: gemini-2.5-flash)
OPENAI_API_KEY	Only if provider is openai	OpenAI API key
DIGEST_AI_MODEL_OPENAI	No	OpenAI model (default: gpt-4o-mini)
OLLAMA_HOST	No	Ollama endpoint (default: http://127.0.0.1:11434)
DIGEST_AI_MODEL_OLLAMA	No	Ollama model (default: llama3.1)
SOURCE_CACHE_TTL	No	Source fetch cache in seconds (default: 3600, 0 disables)
AI_SUMMARY_CACHE_TTL	No	AI summary cache in seconds (default: 86400, 0 disables)
AI_BATCH_DELAY_MS	No	Delay between AI batches in ms (default: 200)

---

Testing

# Unit + feature tests (84 tests, 587 assertions)
composer test

# E2E browser tests (47 tests via Laravel Dusk)
php artisan dusk

---

Project Documentation

• BACKLOG.md — committed work and priorities
• TECH_SPEC.md — architecture, data model, and feature breakdown
• PROGRESS.md — session log and decisions

---

License

MIT