Exploring "smart connections" for note taking

Shreyas Prakash headshot

Shreyas Prakash

Not starting with a blank slate has been a great productivity boost in my writing. I wrote 50K words in 2024. And I can safely say that these 50K words have been written in a well thought manner, instead of an AI generated word salad.

All this, because I’ve been exploring this neat little plugin called as Smart Connections on Obsidian. It is a tool, and I wouldn’t be naive enough to say that tools don’t matter. This tool allows for two main affordances:

  • The ability to ask questions and find answers sourced from your own documents.
  • The ability to make ‘smart connections’ across documents.

The strongest reason why this obscure plugin has changed my life is because of the ability to make semantic search possible.

Why does it matter, and why should we care? Let’s deep dive into this.

Previously if I have to write an article about a ‘dog’, I search for the keyword ‘dog’ across all my notes and then find relevant ones. Absorb information from what I’ve written, and then copy paste snippets into the original doc, remixing and rehashing wherever necessary. However, this method fails to recognise that I might have written articles not just about the keyword ‘dog’, but also it’s close cousins.

![[Attachments/images/2024/12/image-1.png]]

There might be an article on pets which I might draw some inspiration from. It might even be an article on ‘pet parenting’, or ‘pet food’, or ‘dog walking’, ‘pet walking’. A standard keyword search might not provide me these results. And this leads to them getting deeply buried in the existing pile of notes.

This is where semantic search helps. So what’s the key difference between semantic search and keyword search?

Source: Claude website

Let’s say we search in our note-taking app — ‘Where is the world cup’. It pulls in notes based on this query. Some of these top queries include:

  • The world cup is in Qatar.
  • The sky is blue.
  • The bear lives in the woods.
  • An apple is a fruit.

The reason why this was prioritised this way was because the responses have the following number of words in common with the query. The world cup is in Qatar. (4 words in common) The sky is blue. (2 words in common) The bear lives in the woods. (2 words in common) An apple is a fruit. (1 word in common)

Based on this, the winning response is “The world cup is in Qatar”. However, we’re not always lucky. Let’s say we have this search result — ‘Where in the world is my cup of coffee?’. This response has five words in common with the search query — ‘Where is the world cup’. And it would rank first. Although they are not semantically connected.

Semantic search is helpful to find the nearest neighbours.

Why should I care about semantic search when it comes to note taking? With the rise of LLMs and vector based search, it becomes easier and easier for us to find similar documents and pieces of text. We dont have to type in the exact keyword while searching through documents. We just need to be close enough. And the semantic search would do it’s magic and find all the close cousins. You eliminate the cold start problem of writing with a blank slate.

Even for writing this article, as you could see on the right side, I am able to find notes intelligently connected to my current piece.

For example, as I’m writing about ‘smart connections’ in general, it is able to connect this with a tweet by Connected Papers, and a relevant book of interest which talks about the same — The Goldilocks Map. It’s even giving me a confidence level of correlation.

![[Attachments/images/2024/12/shreyas-17-10-2023-at-10.18.04@2x.png]]

The note suggested has 79% similarity with the note I’m writing.

Another use of this plugin is the ability to ask questions based on your internal knowledge (you wouldn’t be able to google this, as this is based on all your personal notes, key insights, wisdom you’ve gained).

You might have already used some of those GPT based PDF chat services. It’s the same, except that this works on all your private notes. For instance:

![[Attachments/images/2024/12/shreyas-17-10-2023-at-10.24.59@2x.png]]

Notes serve as a source of inspiration and material to build upon, helping jumpstart the writing process. This is how I design my writing environment. With a fertile ground of ideas previously explored.

Subscribe to get future posts via email (or grab the RSS feed). 2-3 ideas every month across design and tech

Read more

  1. Hammock driven developmentagentic-coding
  2. Peculiar ways number three fits into our funny little brains
  3. AI sandwich as a defacto principle for anything agentic engineering relatedagentic-coding
  4. How I write essays in 2026writing
  5. Authority in the guise of evidencecritical-rationalism
  6. Map is not the territoryphilosophy
  7. Self hypnosis as a manifestation ritualmeditation
  8. Hegelian dialectic for structured reasoning with AI agentsphilosophy
  9. How I prepare for tough negotiations nowadaysnegotiation
  10. When should we steelthread somethingproduct-development
  11. How to become a polyglot
  12. Breadboarding, shaping, slicing, and steelthreading solutions with AI agentsproduct-management
  13. Healthy conflict in teams have a tipping point
  14. Deslopify AI writing
  15. How I started building softwares with AI agents being non technicalagentic-coding
  16. Read raw transcriptsknowledge
  17. Legible and illegible tasks in organisationsproduct
  18. L2 Fat marker sketchesdesign
  19. Writing as moats for humanswriting
  20. Beauty of second degree probesdecision-making
  21. Boundary objects as the new prototypesprototyping
  22. One way door decisionsproduct
  23. Finished softwares should existproduct
  24. How I periodically rank my rough draftsobsidian
  25. Flipping questions on its headinterviewing
  26. Vibe writing maximswriting
  27. How I blog with Obsidian, Cloudflare, AstroJS, Githubwriting
  28. How I build greenfield apps with AI-assisted codingai-coding
  29. We have been scammed by the Gaussian distribution clubmathematics
  30. Classify incentive problems into stag hunts, and prisoners dilemmasgame-theory
  31. I was wrong about optimal stoppingmathematics
  32. Thinking like a ship
  33. Hyperpersonalised N=1 learningeducation
  34. New mediums for humans to complement superintelligenceai-coding
  35. Maxims for AI assisted codingai-coding
  36. Personal Website Starter Kitai-coding
  37. Virtual bookshelvesaesthetics
  38. It's computational everythingtrends
  39. Public gardens, secret routesdigital-garden
  40. Git way of learning to codeai-coding
  41. Kaomoji generatorsoftware
  42. Copy, Paste and Citeai-coding
  43. Style Transfer in AI writingai-coding
  44. Understanding codebases without using codeai-coding
  45. Vibe coding with Cursorai-coding
  46. Virtuoso Guide for Personal Memory Systemsmemory
  47. Writing in Future Pastwriting
  48. Publish Originally, Syndicate Elsewhereblogging
  49. Poetic License of Designdesign
  50. Idea in the shower, testing before breakfastsoftware
  51. Technology and regulation have a dance of ice and firetechnology
  52. How I ship "stuff"software
  53. Writing is thinkingwriting
  54. Song of Shapes, Words and Pathscreativity
  55. How do we absorb ideas better?knowledge
  56. Read writers who operatewriting
  57. Brew your ideas lazilyideas
  58. Trees, Branches, Twigs and Leaves — Mental Models for Writingwriting
  59. Compound Interest of Private Notesknowledge
  60. Conceptual Compression for LLMsai-coding
  61. Meta-analysis for contradictory research findingsdigital-health
  62. Proof of workproduct
  63. Gauging previous work of new joinees to the teamleadership
  64. Task management for product managersproduct
  65. Beauty of Zettelswriting
  66. Stitching React and Rails togetherai-coding
  67. Exploring "smart connections" for note takingknowledge
  68. Deploying Home Cooked Apps with Railssoftware
  69. Repetitive Copypromptingwriting
  70. Questions to ask every decadejournalling
  71. Balancing work, time and focusproductivity
  72. Hyperlinks are like cashew nutswriting
  73. Brand treatments, Design Systems, Vibesdesign
  74. How to spot human writing on the internetwriting
  75. Can a thought be an algorithm?product
  76. Opportunity Harvestingcareers
  77. How does AI affect UI?design
  78. Everything is a prioritisation problemproduct-management
  79. Nowlifestyle
  80. How I do product roastsproduct
  81. The Modern Startup Stacksoftware
  82. In-person vision transmissionproduct
  83. How might we help children invent for social good?social-design
  84. The meeting before the meetingmeetings
  85. Design that's so bad it's actually gooddesign
  86. Lessons learnt interview prepping for product rolesinterviewing
  87. Obsessing over personal websitessoftware
  88. English is the hot new programming languagesoftware
  89. Better way to think about conflictsconflict-management
  90. The role of taste in building productsdesign
  91. Dear enterprises, we're tired of your subscriptionssoftware
  92. Products need not be user centereddesign
  93. World's most ancient public health problemsoftware
  94. Pluginisation of Modern Softwaredesign
  95. Let's make every work 'strategic'consulting
  96. Making Nielsen's heuristics more digestibledesign
  97. Startups are a fertile ground for risk takingentrepreneurship
  98. Insights are not just a salad of factsdesign
  99. Minimum Lovable Productproduct
  100. Methods are lifejackets not straight jacketsmethodology
  101. How to arrive at on-brand colours?design
  102. Minto principle for writing memoswriting
  103. Importance of Whytask-management
  104. Quality Ideas Trump Executionsoftware
  105. How to hire a personal doctor
  106. Why I prefer indie softwareslifestyle
  107. Use code only if no code failscode
  108. Self Marketing
  109. Personal Observation Techniquesdesign
  110. Design is a confusing worddesign
  111. A Primer to Service Design Blueprintsdesign
  112. Rapid Journey Prototypingdesign
  113. Visualise detailed file structures on CLIcli
  114. Do's and Don'ts of User Researchdesign
  115. Design Manifestodesign
  116. Complex project management for productproducts
  117. How might we enable patients and caregivers to overcome preventable health conditions?digital-health
  118. Pedagogy of the Uncharted — What for, and Where to?education
  119. Future of Ageing with Mehdi Yacoubiinterviewing
  120. Future of Tacit knowledge with Celeste Volpiinterviewing
  121. Future of Rural Innovation with Thabiso Blak Mashabainterviewing
  122. Future of Equity with Ludovick Petersinterviewing
  123. Future of work with Laetitia Vitaudinterviewing
  124. Future of Mental Health with Kavya Raointerviewing
  125. Future of unschooling with Che Vanniinterviewing
  126. How might we prevent acquired infections in hospitals?digital-health
  127. The why to endure any howentrepreneurship
  128. Design education amidst social tribulationsdesign
  129. How might we assist deafblind runners to navigate?social-design