Discovery Synthesis Creation The Research Workspace

Research is curiosity. Curiosity is human. So research should feel human too—a natural extension of how you think and work.

Meet

Ada

The AI-native research workspace for deep thinkers.

Join the Beta See how it works →

Research is a living system.

Traditional tools fragment your thinking—papers in one place, annotations in another, writing somewhere else, and your ideas scattered everywhere.

Ada keeps everything connected in one living workspace, so context carries forward and your thinking compounds over time.

One workspace. One evolving context. One connected thread of thought.

LIBRARY READER NOTES BROWSER
arxiv.org/list/cs.CL/recent
|
Library
Attention Paper New
BERT Notes
GPT-3 Analysis
Reader
Notes
Canvas

Attention Is All You Need

Vaswani, Shazeer, Parmar, et al. · NeurIPS 2017

Abstract

The dominant sequence transduction models are based on complex recurrent or convolutional neural networks. The best performing models also connect the encoder and decoder through an attention mechanism.

1. Introduction

Recurrent neural networks, long short-term memory and gated recurrent neural networks in particular, have been firmly established as state of the art approaches.

Notes & Highlights
Highlight

"The best performing models also connect the encoder and decoder through an attention mechanism."

Page 1, Abstract

Note

Key insight: Attention replaces recurrence entirely → enables parallelization

Linked to [[Transformers]]

Question

How does this compare to BERT's bidirectional approach?

Ada Ready
What's the key innovation here?
The Transformer replaces recurrence with self-attention, enabling O(1) sequential ops and full parallelization.

Built for students, scholars, builders, and independent thinkers.

Not just a tool. A research protocol.

Pick your deliverable. Ada runs the workflow.

Nine deliverables. Each with its own protocol.

Ada doesn't treat every project the same. Pick your deliverable type, answer context questions, and the system configures databases, quality frameworks, extraction columns, and citation style automatically.

  • Systematic reviews to white papers
  • Context questions adapt each workflow
  • Community templates from published researchers

Database-specific search. Not one generic query.

PubMed uses MeSH terms. PsycINFO uses Thesaurus. ERIC uses Descriptors. Ada generates the right Boolean syntax for each database, logs every search for reproducibility, and tracks yield per source.

  • PICO-structured keyword tables
  • PubMed, ERIC, PsycINFO, Cochrane, JSTOR
  • Timestamped search audit trail

Structured extraction. Every value source-linked.

For every included study, Ada reads the full paper and extracts data into the Matrix. Each cell links back to the exact page and paragraph. Columns auto-configure by field: education gets grade level and WWC rating, psychology gets effect sizes and measurement tools.

  • Field-specific column configurations
  • Source-linked to PDF page & paragraph
  • Confidence indicators on uncertain data

The right framework for each study design.

RoB 2 for randomized trials. Newcastle-Ottawa for observational studies. JBI for qualitative. Ada rates each bias domain, generates traffic-light plots, and auto-fills PRISMA 2020 flow diagrams from your actual screening data.

  • RoB 2, Newcastle-Ottawa, JBI, GRADE
  • Auto-generated PRISMA 2020 diagrams
  • Reporting checklists pre-filled from workflow

Every section drafted. Every claim linked to evidence.

Ada drafts structured abstracts, methods with full search strategy detail, results organized by theme, and discussion sections connecting your findings to prior work. Mock review scoring before you submit.

  • Section-by-section with inline citations
  • PRISMA checklists auto-filled with page numbers
  • Mock study section review before submission
Protocol
@ Sources
Standards
Choose your deliverable
Systematic Review
Journal Article
Thesis
Conference Paper
Grant Application
Research Proposal
Fellowship
Policy Brief
White Paper
Systematic Review
PRISMA 2020 · Exhaustive search · Quality appraisal
01
Question Formulation
PICO / PEO / PCC framework
02
Search Strategy
Keywords, synonyms, Boolean strings
03
Literature Search
Multi-database execution + dedup
04
Screening
Title-abstract + full-text with AI
05
Data Extraction
Structured Matrix with source linking
06
Quality Appraisal
RoB 2, Newcastle-Ottawa, JBI
07
Synthesis & Drafting
Narrative synthesis + PRISMA report
Keyword / Synonym Table Field: Education
Concept Primary Terms Synonyms & Variants
P
K-12 students
pupils, adolescents, children, youth, "school-age"
I
project-based learning
PBL, inquiry-based, experiential learning
O
academic achievement
test scores, GPA, learning outcomes, performance
ERIC
DE "Project Based Learning" AND DE "Academic Achievement" AND DE "Elementary Secondary Education"
PubMed
("project-based learning"[tiab] OR "inquiry-based"[tiab]) AND ("academic achievement"[tiab] OR "learning outcomes"[tiab]) AND (child*[tiab] OR adolescen*[tiab])
PsycINFO
MAP "Experiential Learning" AND MAP "Academic Achievement" AND AG "school age (6-12)" OR AG "adolescence (13-17)"
ERIC342
PubMed187
PsycINFO156
After dedup489
Evidence Matrix Config: Education (IES Standards) 0 / 23 studies extracted
Author Year Design N Grade Effect Size WWC
Williams & Chen 2021 RCT 248 6-8 0.42 Meets
Krajcik et al. 2019 Quasi-exp 412 9-10 0.31 Meets w/ res.
Thomas 2020 RCT 186 3-5 0.55 Meets
Barak & Dori 2022 Quasi-exp 315 7-8 0.28 ? Does not meet
Han et al. 2023 RCT 520 K-2 0.38 Meets
Risk of Bias Assessment Framework: RoB 2 (Cochrane)
Study D1
Random. D2
Deviat. D3
Missing D4
Measur. D5
Report. Overall
Williams & Chen (2021)
Krajcik et al. (2019)
Thomas (2020)
Barak & Dori (2022)
Han et al. (2023)
Low risk Some concerns High risk
PRISMA 2020
Identified
685
↓
Screened
489
↓
Included
23
Sections
Abstract
Introduction
Methods
Results
Discussion
PRISMA Checklist
Title p.1
Abstract p.1
Eligibility p.4
Search strategy
Selection
Methods 1,247 words
2.1 Search Strategy
A systematic search was conducted across three electronic databases: ERIC, PubMed, and PsycINFO. Search strategies were developed using database-specific controlled vocabulary and free-text terms.
For ERIC, Descriptors including "Project Based Learning" and "Academic Achievement" were combined with free-text terms. For PubMed, Medical Subject Headings (MeSH) were used where applicable (see Appendix A for full search strings).
The search yielded 685 records. After removing 196 duplicates, 489 unique records were screened at the title-abstract level. Of these, 67 proceeded to full-text review, and 23 studies met all inclusion criteria (Figure 1).
2.2 Quality Assessment
Risk of bias was assessed using the Cochrane RoB 2 tool for randomized trials (Sterne et al., 2019) and the Newcastle-Ottawa Scale for quasi-experimental designs (Wells et al., 2000).

For the curious.

Ada is for the way research actually happens — nonlinear, iterative, and driven by curiosity.

Whether you're writing a thesis, synthesizing literature, building a new argument, or exploring an unfamiliar field, Ada keeps everything connected so your work stays coherent over time.

Built for Thesis Writers 200+ sources, long timelines
Built with researchers from
Join the Future

Research, without limits.

Ada is being built alongside researchers, academics, and thinkers at institutions like Harvard, Stanford, MIT and many more to redefine how we interact with information.

Free during beta · macOS first · Windows coming soon