Overview
StudyLens gives non-experts the same critical evaluation framework as an experienced science communicator. Users paste a DOI, PMID, or study title and get a complete analysis: what the study actually found, why they should be cautious about the findings, how media covered it, and the context they need to evaluate evidence on their own.
- Source-cited extraction: Every fact links to its exact source sentence in the paper—no black-box summaries
- Methodological Caution Meter: A composite 0–100 score surfacing p-hacking risk, surrogate endpoints, causal overclaims, and more
- Media accuracy analysis: Automated discovery of news coverage with stance classification and accuracy scoring
- Teaching layer: Contextual cards that explain every flag so users learn to recognize issues themselves
Problem
Every week, nutrition studies make headlines. The public reads them and makes decisions—what to eat, what supplements to buy, what to worry about. Almost none of these headlines accurately represent the research.
Misleading headlines
- Observational studies reported as proof of causation
- Relative risks without absolute risks
- Surrogate endpoints presented as clinical outcomes
Hidden methodology
- Negative primary outcomes buried
- Industry funding undisclosed
- Post-hoc subgroup findings headlined
No tools for non-experts
- Existing tools target researchers, not the public
- No product combines extraction + quality + media + education
- 99% of people never learned what to look for
Solution
StudyLens automates the analytical framework of an experienced science communicator and makes it available to everyone in under 15 seconds.
How it works
- Paste a study identifier: DOI, PMID, title, or URL—StudyLens resolves and fetches the paper from open-access sources
- Automated extraction: Claude extracts 16 fact categories with source citations and confidence scores
- Interrogation flags: 13 methodological flags across 4 tiers surface what a skeptic would notice
- Media discovery: An LLM agent finds, validates, and classifies news coverage automatically
- Evidence landscape: Meta-topic pages show where the study fits in the broader body of research
Architecture
A modern full-stack application with an event-driven extraction pipeline, serverless Postgres, and infrastructure as code.
Application layer
- Next.js 16 + React 19: App Router with TypeScript strict mode
- Tailwind CSS v4: Design tokens with full dark-mode support
- Auth.js v5: Google OAuth with session-based access control
- Drizzle ORM: Type-safe queries with migration generation
Infrastructure
- Neon PostgreSQL: Serverless Postgres with branching
- Inngest: Event-driven durable functions for extraction pipeline
- Terraform: Vercel + Neon infrastructure as code
- Docker Compose: Local development environment
Extraction pipeline
| Step | What happens |
|---|---|
| 1. Resolve identifiers | OpenAlex (primary), Crossref, PubMed, Semantic Scholar for maximum metadata coverage |
| 2. Check OA status | Unpaywall API determines open-access availability and best source URL |
| 3. Fetch full text | PMC XML preferred, OA location fallback, DOI landing page as last resort |
| 4. Extract + interrogate | Claude structured tool calling: 16 fact categories, 13 interrogation flags, topic classification |
| 5. Store & classify | Study row, facts, flags, topic associations, coverage badge (gold/silver/bronze) |
| 6. Media discovery | LLM agent searches, validates, and classifies news coverage with stance analysis |
The four layers
StudyLens delivers value through four complementary layers, each building on the last.
Layer 1: Extraction — “What did the study actually find?”
- 16 fact categories: study type, sample size, outcomes, effect sizes, funding, blinding, and more
- Every claim sourced to a specific sentence with section and character offset
- Confidence scores for each extraction
- Coverage badges: gold (full text), silver (abstract), bronze (metadata only)
Layer 2: Interrogation — “Should I believe it?”
- 13 flags across 4 tiers: statistical red flags, effect size analysis, study design adequacy, transparency
- Composite Methodological Caution Meter (0–100)
- P-hacking detection: outcome switching, subgroup mining, borderline significance
- Surrogate endpoint flagging, causal overclaims, funding independence
Layer 3: Context — “How does the world see it?”
- Automated media discovery agent: search, validate, classify coverage
- Stance classification: positive, negative, cautious, neutral
- Accuracy scoring per article
- Meta-topic pages with LLM-generated evidence summaries
Layer 4: Education — “Why does this matter?”
- Teaching cards for every flag and methodology concept
- Consistent structure: what is it, why it matters, what to look for
- Contextual display—only shows cards relevant to the current study
- Goal: users outgrow the tool by building their own evidence literacy
Outcomes
<15s
DOI to full analysis
$0.04
Cost per study analyzed
16
Fact categories extracted
14
Design documents
Delivered capabilities
- 5-step extraction pipeline with idempotent Inngest steps
- 16 fact categories + 13 interrogation flags via Claude structured tool calling
- Deduplication across DOI, PMID, PMCID, and title
- Google OAuth with session-based access control
- Collections system for organizing saved studies
- Full dark-mode design system with semantic CSS tokens
- Freemium revenue model with 5 pricing tiers
- Stripe billing with webhook-driven entitlements
- Docker + Terraform infrastructure as code
- WCAG AA accessible with mobile-first responsive layouts
Building an AI-powered content platform?
StudyLens demonstrates end-to-end delivery of an LLM extraction pipeline, event-driven architecture, and consumer-facing SaaS with freemium billing. If you're tackling similar challenges, let's talk.