<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Sandip Dev on Medium]]></title>
        <description><![CDATA[Stories by Sandip Dev on Medium]]></description>
        <link>https://medium.com/@sandipdev?source=rss-9cf403e37e8e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Z90UPK4c4jMUUaxOzmUjYA.png</url>
            <title>Stories by Sandip Dev on Medium</title>
            <link>https://medium.com/@sandipdev?source=rss-9cf403e37e8e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 05:19:25 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@sandipdev/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[An Easy-to-Build, Easy-to-Manage AI PM OS: 8 Skills, 3 Triggers, 7 Scheduled Jobs]]></title>
            <link>https://sandipdev.medium.com/an-easy-to-build-easy-to-manage-ai-pm-os-8-skills-3-triggers-7-scheduled-jobs-c1df988e4c4d?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/c1df988e4c4d</guid>
            <category><![CDATA[aio]]></category>
            <category><![CDATA[aipm]]></category>
            <category><![CDATA[ai-product-management]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[ai-pm-os]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Sat, 30 May 2026 17:46:21 GMT</pubDate>
            <atom:updated>2026-05-30T17:46:21.026Z</atom:updated>
            <content:encoded><![CDATA[<p>I’ve been using Claude Code as my main interface for about six months. I build with it, I run my day through it, I apply for jobs with it, I prep for system design interviews with it. Four jobs, one tool, sometimes inside the same hour.</p><p>The frustrations were specific. I lost track of decisions between sessions. I logged the same thing twice on the same day because Claude didn’t remember the morning. I said “I’ll remember this” about meeting takeaways and then didn’t. I knew where my notes <em>should</em> go and still skipped the filing because the routing was a 30-second prompt cascade that killed the moment.</p><p>The whole system is called Samaritan. (I’m a Person of Interest fan — yes, named after the show’s antagonist AI. The version I built has the same ambition the show’s did, but the opposite scope: it watches one person, by request, on purpose. Repo: <a href="https://github.com/devsandip/samaritan">github.com/devsandip/samaritan</a>.)</p><p>The second-brain post I published earlier this week covers the architecture I built underneath — Obsidian for prose, Notion for structured rows, Claude Code as the broker. This post is the layer on top. The operating system. What I actually do every day.</p><p>It’s eight skills, a trigger model with three rails of coverage, seven scheduled jobs, and one Cowork dashboard. The whole thing took about three weekends. Most of it is Claude Code plugin skills and one MCP scheduled-tasks config. No custom infrastructure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Mj3yWMyF8DP4xkDH.png" /></figure><h3>Eight skills, one job each</h3><p>Capture is a habit problem, not a tooling problem. The trick is making the friction-per-capture so low that the habit survives a bad week.</p><p>I have eight skills, installed as a single Claude Code plugin. Each one maps to a slash command and a single, narrow job. The one in bold is the load-bearing one — if I had to keep one and lose the other seven, I’d keep it.</p><p><strong>/wrapSession-close - scan the conversation, file everything to the right place, report backSilent approval. Auto-fires on session-end phrases. &quot;Undo last wrap&quot; within 24h.</strong></p><p>7-day digest written to Areas/Weekly/YYYY-Www.md</p><p>Computes time allocation from the hourly log. Falls back to calendar + TickTick when the log was sparse.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*5mrmRfQACsgCQkPt.png" /></figure><h3>/wrap is the one that holds it together</h3><p>Most personal knowledge systems die the same way. You capture for a month, you stop wrapping things up at the end of sessions, and three weeks later you have a folder full of half-formed notes and no idea what’s filed, what’s pending, and what you’ve already decided. The system silently calcifies.</p><p>/wrap solves this by running on the residue of any conversation. At session end - when I type &quot;we&#39;re done&quot; or &quot;thanks, let&#39;s wrap&quot; or just &quot;ok bye&quot; - the skill scans the chat for decisions made, insights captured, people mentioned, next-step tasks, and files them all. Silently. No per-item approval cascade.</p><p>Silent approval is the controversial choice. Most AI-agent guidance says you should confirm before any write. I disagree, sharply, for one specific reason: per-item approval kills the habit. If /wrap triggers a 30-second prompt cascade - &quot;file this decision? Y/N. Add this insight? Y/N.&quot; - I will not wrap, and the system rots. The failure mode of over-wrapping is recoverable; the failure mode of under-wrapping is data loss.</p><p>The safety net is “undo last wrap” within 24 hours. If Claude got something wrong, I can roll it back the next morning. In four weeks of use I’ve used it twice. Both times the row was wrong, not the skill — Claude had filed something I’d thought-out-loud-and-rejected as a decision. Two false positives in ~40 wraps is well inside my tolerance.</p><p>If you build one skill of this set, build /wrap. The rest are nice. This one is load-bearing.</p><h3>Three rails, one truth: nothing falls through</h3><p>Every captured event has at least two of three triggers covering it. Belt-and-suspenders by design.</p><ol><li><strong>Live capture</strong> — I explicitly invoke a skill (/log, /decision, /capture, etc.). Highest fidelity. Requires habit.</li><li><strong>Source sweep</strong> — a scheduled job reads a source I didn’t explicitly mention (Gmail, Fireflies, calendar, TickTick completions). Catches what I forgot.</li><li><strong>Session close</strong> — /wrap at the end of any conversation. Picks up the residue.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Ppb0rWVt6HBjBPgH.png" /></figure><p>The redundancy is the trick. Single-trigger systems are fragile — one bad week and you have holes. With three rails, any single rail going dark for a few days is recoverable from the other two. The hourly log on a busy day might be sparse, but the nightly sweep reconstructs it from calendar + TickTick. The meeting I forgot to summarise gets caught by the Fireflies post-meeting sweep. The decision I made in chat but didn’t file gets caught by /wrap.</p><p>This is the most-borrowed and least-discussed idea in the system, so I’ll call out where I took it from: it’s the same principle as defense-in-depth in security — assume any one layer will be bypassed, design for the redundancy to catch it. Old idea, new domain.</p><h3>/log from anywhere: Claude Code Channels + Telegram</h3><p>The biggest gap in most personal knowledge systems is the moment of capture when you’re not at your desk. I solved it with Anthropic’s Claude Code Channels — the Telegram-bridge feature that lets a bot forward messages into a running Claude Code session.</p><p>Here’s the flow:</p><pre>Phone (Telegram) Cowork session Vault + Notion ───────────────── ──────────────── ────────────── Send: &quot;/log just Channels MCP receives, Claude resolves date, finished pricing ───────► forwards to running ────► infers project from review with Sarah&quot; Claude session recent context. Appends to Areas/Daily/ 2026-05-30.md under ## Hourly Log See: &quot;Logged ◄────────── Reply: &quot;Logged 14:00 ◄──── Confirmation 14:00 → [[Onboarding → [[Onboarding Redesign]]&quot; Redesign]]&quot;</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*jL9YVjyXl920xTmC.png" /></figure><p>The Telegram bot is an MCP server. When I message the bot, it forwards into the Claude Code session running on my Mac. Claude processes it with the session’s full local environment — vault access, Notion, TickTick, calendar. It’s not a cloud-hosted agent. The Mac is the agent. Telegram is just a window into it.</p><p>Why this matters for the hourly log specifically: the hourly log is the highest-fidelity signal in the system, granular enough to reconstruct what actually happened in a week, sparse enough to be maintainable. Without Telegram routing, /log would only fire when I&#39;m at my desk - which means lunch hours and afternoon meetings get logged as a clump or not at all. With Telegram routing, &quot;just finished standup&quot; gets captured the moment it happens.</p><p>The nightly inbox processor at 22:00 handles the days I forget entirely. Claude reads calendar + TickTick completions + chat history and proposes draft entries for missed hours. I approve them in the morning. Telegram + nightly reconstruction is what makes the hourly log a survivable habit instead of a six-week experiment.</p><p>Cost: at eight to ten logs per day, the Telegram traffic is pennies a month.</p><h3>Seven scheduled jobs that turn capture into thinking</h3><p>Capture without synthesis is hoarding. The cron jobs are where the system earns its keep.</p><p>Post-meeting sweep</p><p>~30 min after each meeting</p><p>Pulls latest Fireflies transcript, runs the /meeting flow automatically.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*OlCSFw3joXHfWlrk.png" /></figure><p>A few details that matter more than they look:</p><p>The morning briefing goes to <strong>both</strong> Obsidian and Telegram on purpose. Obsidian holds the full briefing in today’s daily note (auditable, history). Telegram pushes a condensed version (glanceable). Mini-briefings are Telegram-only — they’re meant to be ephemeral.</p><p>The mini-briefings skip when nothing has changed. This is the difference between a useful nudge and a notification you mute after two days. If three hours go by and no decisions were made, no meetings happened, no new tasks landed, the 14:00 ping just doesn’t fire.</p><p>The stale check <strong>proposes</strong> archive moves, never auto-archives. A 30-day quiet project is a flag, not a verdict. Often that’s a backburner thing I’m still going to ship. The system raises its hand; I decide.</p><p>Token cost across all jobs runs maybe $2–3 per day. The weekly rollup is the only one that’s even close to expensive, and it’s worth its weight three times over.</p><h3>The dashboard, in one frame</h3><p>The morning answer to “what’s the state of my world right now” used to take three tabs. Now it’s one Cowork artifact.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*nwMnCo65SzHAi24l.png" /></figure><p>It pulls live from TickTick (today’s tasks, this week’s tasks, distribution by project) and Notion (active projects, recent decisions, projects needing attention, open questions). One HTML file, ~400 lines, three whitelisted CDN scripts. Persists across sessions. Reloads pull fresh data.</p><p>I look at it twice a day — first thing in the morning and around 5pm before I close out. It has replaced two browser tabs and one app icon.</p><p>This is also where the build had its longest debugging arc. The Cowork artifact sandbox returns MCP responses wrapped in the MCP Content protocol envelope, not raw — {content: [{type: &quot;text&quot;, text: &quot;&lt;json string&gt;&quot;}]}. Every call appeared to succeed but every data array came back empty. Three iterations to figure out that I needed a defensive unwrap that JSON-parses the inner text. Lesson kept for next time.</p><h3>Three files, one contract</h3><p>The system is contract-driven. Claude reads a small, explicit set of routing rules at session start and behaves according to them — instead of re-deriving where things go each time.</p><p>Three files, three different readers:</p><ul><li><strong>CLAUDE.md</strong> — 1-page orientation. Loaded into every Claude Code session start.</li><li><strong>AGENT_OS.md</strong> — the full operating contract. Routing rules, scheduled jobs, Notion data source IDs. Capped at 3 pages — a manual longer than that is a sign the system is too complex.</li><li><strong>README.md</strong> — human cheat sheet. I read this. I never read the contract.</li></ul><p>Splitting by reader and cadence keeps each one tight. CLAUDE.md changes monthly. AGENT_OS.md changes when a routing rule changes. README.md changes when I need a reminder of my own conventions.</p><h3>What this is sitting on</h3><p>Three substrate pieces did most of the heavy lifting:</p><ul><li><strong>Claude Code</strong> — the agent layer. The reason this took weekends instead of a quarter is that the broker isn’t custom-coded glue, it’s a general-purpose AI assistant reading a contract file.</li><li><strong>Cowork artifacts</strong> — persistent HTML pages with MCP access. The dashboard wouldn’t exist without this; it’s the only surface that sees Notion + TickTick + (eventually) Calendar in one frame and survives session restarts.</li><li><strong>Claude Code Channels</strong> — the Telegram bridge. Without it, /log only fires at the desk. With it, the moment of capture is anywhere I have my phone.</li></ul><p>Honest credit to the apps doing the actual work — <strong>TickTick</strong> for tasks, <strong>Notion</strong> for the structured store, <strong>Obsidian</strong> for the prose store. Claude is the broker. They’re the substrate.</p><h3>What it costs and what it doesn’t</h3><p>Real costs:</p><ul><li>Anthropic API usage for Claude Code: I’m on Claude Max, no marginal cost</li><li>Notion: paid plan, $10/month, would have it anyway</li><li>TickTick: premium, $3/month, would have it anyway</li><li>Obsidian: free</li><li>Cowork plugin: free</li><li>Telegram bot: free</li></ul><p>Total marginal cost of running this system on top of subscriptions I already pay for: <strong>zero</strong>.</p><p>What it doesn’t cost: time I don’t have. The capture skills are 5–10 seconds each. /wrap is 30 seconds at session end and saves me 10 minutes of &quot;wait, where did I put that&quot; the next morning. Net positive every day.</p><h3>What’s not built yet, honestly</h3><p>I’m posting this at four weeks of use, not four years. Two things are designed but not running:</p><ul><li>The seven scheduled jobs exist as specs in AGENT_OS.md. I have not wired them up via the scheduled-tasks MCP yet. The morning briefing and weekly rollup are the next concrete build step.</li><li>The Telegram Claude Code Channels bot is designed. Not configured yet. /log from the desk works; /log from the phone is the next milestone.</li></ul><p>The reason I’m publishing anyway is that the architecture is the part worth sharing. The wiring is a weekend.</p><h3>Credit where it’s due</h3><ul><li><strong>Anthropic</strong> for Claude Code, Cowork, and Claude Code Channels. The substrate that made this buildable in weekends. Three months ago I would have written custom integrations for each app; today the agent layer is general-purpose and the integrations are MCPs.</li><li><strong>TickTick, Notion, Obsidian</strong> — the apps doing the actual work. Claude is the broker. They’re the substrate underneath.</li><li><strong>Tiago Forte</strong> — the PARA vocabulary I’m still using. <em>Building a Second Brain</em> is in the bibliography for the architecture-level post that precedes this one.</li><li><strong>The Claude Code + Obsidian writers</strong> on Medium and Substack — the running conversation that made this feel like a viable build path rather than a frontier experiment.</li></ul><h3>Try it. Tell me what breaks.</h3><p>The plugin, the AGENT_OS.md contract, the four Obsidian templates, the Cowork dashboard, and these three blog posts as docs are all in the repo: <a href="https://github.com/devsandip/samaritan">github.com/devsandip/samaritan</a>. Fork it, gut it, adapt it — that’s the point of shipping it open.</p><p>What I most want is failure reports. The skills are deterministic — they file or they don’t — but the routing logic in AGENT_OS.md is opinionated and the opinions are mine. If your work shape doesn’t fit, I want to know which contract assumption broke.</p><p>If you’ve built something in this space and it survived more than three months of real use, I’d love to hear what you kept and what you cut.</p><p><em>Originally published at </em><a href="https://www.blog.sandip.dev/2026/05/30/ai-pm-os-skills-triggers-jobs/"><em>https://www.blog.sandip.dev</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c1df988e4c4d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A PM Building a Second Brain: Why I Skipped the Vector DB and Picked Claude + Notion + Obsidian]]></title>
            <link>https://sandipdev.medium.com/a-pm-building-a-second-brain-why-i-skipped-the-vector-db-and-picked-claude-notion-obsidian-32b69af7d752?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/32b69af7d752</guid>
            <category><![CDATA[aipm]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[ai-product-management]]></category>
            <category><![CDATA[aio]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Sat, 30 May 2026 17:44:58 GMT</pubDate>
            <atom:updated>2026-05-30T17:44:58.207Z</atom:updated>
            <content:encoded><![CDATA[<p>I’ve been using Claude Code as my main interface for about six months now. I build with it, I run my day through it, I apply for jobs with it, I prep for system design interviews with it. Four different jobs, one tool, sometimes inside the same hour.</p><p>Two patterns kept showing up.</p><p>The first was repeat work. I’d have the same conversation twice in a week — about a decision I’d already made, a person I’d already met, a tradeoff I’d already worked through — because Claude didn’t remember and I didn’t have anywhere durable to put it. The second was lost context. Every new session started cold. Every fresh window was a clean slate. The “ChatGPT but for me” promise of personalised memory was still mostly a promise.</p><p>So I built a second brain. Not the “weekend vector database” version. A real one — capture, store, retrieve, synthesise — with Claude as the broker between three apps I already used. I’m a Person of Interest fan, so I called it Samaritan. The joke being: this one only watches one person on purpose — me. Repo: <a href="https://github.com/devsandip/samaritan">github.com/devsandip/samaritan</a>.</p><p>This is the writeup of how I chose the architecture, what I rejected, and why.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*J0BirP8J0H9g3Dov.png" /></figure><h3>The five architectures I considered</h3><p>Before building anything, I spent a weekend mapping out what other people had already done. The Personal Knowledge Management space is crowded — Tiago Forte’s PARA and <em>Building a Second Brain</em> are the canonical reference, Brian Petro’s Smart Connections is the Obsidian-native semantic search plugin, Khoj.dev (YC W24) is the self-hosted-AI-second-brain option, and there’s a growing Claude Code + Obsidian writer ecosystem on Medium and Substack publishing every variant of “I plugged Claude into my notes.”</p><p>I scored five plausible architectures on five axes: ease of build, fit for PM work, novelty, external dependencies, and token cost.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*W0pF9thEi_GgV1kk.png" /></figure><p>A</p><p>Pure Obsidian + Cowork file access</p><p>Baseline. Works today. Ceiling is low.</p><p>B</p><p>Auto-memory + Obsidian sync skill</p><p>Underrated. Flashy enough that no one writes about it.</p><p>C</p><p>Local vector DB (Smart Connections / Khoj / custom Chroma MCP)</p><p>Coolest. Highest infrastructure burden.</p><p>D</p><p>Obsidian (prose) + Notion DBs (structured) + Claude broker</p><p>Cleanest map onto entity-heavy PM work.</p><p>E</p><p>Daily-notes pipeline + scheduled synthesis</p><p>Highest ceiling. The system improves the system.</p><p>I picked <strong>D + E combined</strong>. Notion gives me the structured backbone PM work actually needs — projects, decisions, people, insights as queryable rows. Scheduled synthesis converts captured signal into thinking. Obsidian holds the prose. Claude brokers between them.</p><p>I skipped C — the vector database route — deliberately. That decision deserves its own section, because it’s the one most people get wrong.</p><h3>Why I skipped the vector database</h3><p>The default advice for “build yourself an AI second brain” right now is: install a vector database. Either Smart Connections inside Obsidian, or Khoj as a separate app, or roll your own Chroma + embedding model + MCP server. Retrieval-augmented generation is the standard pattern, and there’s a year of writing online telling you to do it.</p><p>For a PM’s note volume, it’s the wrong tool.</p><p>Two specific reasons. First, my notes are entity-heavy, not text-heavy. The things I actually need to retrieve are <em>Decision DEC-12</em>, <em>Project Onboarding Redesign</em>, <em>Person Sarah at Acme</em> — structured rows with relations. Vector similarity is a fuzzy text match. A Notion database query is an exact one. For “what did we decide about pricing last quarter?”, a structured query against a Decisions table beats a semantic search every time.</p><p>Second, the infrastructure cost is real and the benefit is speculative. A vector DB means embedding every note on write, hosting the index somewhere, keeping it in sync, and tuning chunk size and overlap until the recall feels reasonable. For a few thousand notes, that’s a weekend of setup and ongoing care. Structured retrieval — open the right Notion database, filter on a property, follow a relation — is one MCP call and instant.</p><p>The mechanism matters. Vector search is correct when your data is unstructured and your queries are about meaning. PM data is mostly structured and the queries are mostly about specific entities. Picking the right retrieval style is upstream of every other choice.</p><p>I’ll revisit if I ever hit a real recall problem. I haven’t yet.</p><h3>The four-store ownership rule</h3><p>Once I picked D + E, the hard part wasn’t picking tools. It was deciding what lived where, and refusing to duplicate. Every “AI second brain” failure I’ve seen comes from the same place: the same fact gets written into two stores, they drift, and now you don’t know which one is true.</p><p>So I wrote an ownership rule. Each store owns one thing. No overlap.</p><p><strong>Obsidian</strong></p><p>Prose, drafts, meeting notes, daily notes, hourly log, research synthesis</p><p>Structured rows, tasks</p><p><strong>Notion</strong></p><p>Projects, Decisions, People, Insights — queryable rows with relations</p><p>Long-form writing, drafts</p><p><strong>TickTick</strong></p><p>Tasks with due dates, recurring actions, next-steps owned by me</p><p>Notes, decisions, references</p><p><strong>Claude memory</strong></p><p>Cross-session preferences, working facts about me</p><p>Time-series data, anything queryable</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yJc3nXL7833KXfWU.png" /></figure><p>The mental model that holds it together: Obsidian is the journal, Notion is the CRM, TickTick is the task list. Together they feel like one system. Claude is what makes them feel that way.</p><p>The TickTick separation is the one people argue with most. “Just use Notion for tasks.” No. TickTick is purpose-built for tasks — mobile-first, recurring, notifications, swipe-to-complete. Replicating that inside Notion would be worse on every dimension. The cost of a third store is one ID mapping inside the Notion Projects row. That’s it.</p><h3>Four Notion databases, and a rule for resisting a fifth</h3><p>Most PM second-brain templates accumulate databases. Tasks, notes, areas, goals, OKRs, weekly reviews, monthly reviews. Each extra database doubles routing complexity — every capture moment becomes a “which one of these does this go in?” question, and that’s the question that kills the habit.</p><p>I held to four:</p><ul><li><strong>Projects</strong> (hub — everything relates back here)</li><li><strong>Decisions</strong> (audit trail — what was decided, why, by whom, reversible or not)</li><li><strong>People</strong> (lightweight CRM — name, role, last interaction, related projects)</li><li><strong>Insights</strong> (catch-all reference for everything else worth keeping)</li></ul><p>Tasks live in TickTick. Areas live as flat folders and tags in Obsidian.</p><p>The threshold for adding a fifth: a query I genuinely can’t answer otherwise. Not “it would be nice to have.” Not “Notion has a template for it.” A real, repeated, unanswerable query. I haven’t hit it yet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NCHfjGdvvt5drftL.png" /></figure><h3>Flat folders, not nested PARA</h3><p>Tiago Forte’s PARA — Projects, Areas, Resources, Archives — is the most-cited filing system in the personal knowledge management world. I use it. With one departure from the canonical version: I keep folders flat.</p><p>PARA’s strict mutual exclusivity rule — every note lives in exactly one of the four categories, and you move it as its status changes — is a paper-folder constraint. It came from an era when a note physically had to be in one drawer. Obsidian’s wikilinks and tags carry multi-belonging natively. A note can live in Projects/Onboarding/ and still be linked from three project pages, two area pages, and a daily note. Folders represent &quot;where I&#39;m acting on this.&quot; Tags and links represent &quot;what this relates to.&quot;</p><p>The win: when an AI is reading your vault, multi-belonging is what lets it answer cross-cutting questions. “What did I work on last month that touched onboarding?” doesn’t require a folder hierarchy. It requires links and tags. The folder is just the parking spot.</p><p>I’m not the first person to land here — every Obsidian power user eventually does — but I’m calling it out because PARA is so widely cited that the nested version still gets recommended, and it’s wrong for the LLM era.</p><h3>What everything sits on</h3><p>Three substrate choices made this buildable in a weekend instead of a quarter:</p><ul><li><strong>Claude Code</strong> as the agent, with MCP servers for Notion, TickTick, and the filesystem. The fact that the broker layer is a general-purpose AI assistant — not a custom-coded integration — is what makes the system bend to my workflow instead of the other way around.</li><li><strong>Obsidian</strong> as the prose store. Markdown files on disk. Plain text. Mine forever, regardless of what happens to any company.</li><li><strong>Notion</strong> as the structured store. Worth the lock-in cost because the database UX is unmatched and the MCP is solid.</li></ul><p>The combination is the point. Claude alone is amnesiac. Obsidian alone has no structured query layer. Notion alone has no prose ergonomics. Glue them together with an opinionated broker and you get a second brain that does what no single one of them can.</p><h3>What I’m crediting</h3><p>A few people whose thinking I borrowed from while building this:</p><ul><li><strong>Tiago Forte</strong> — <em>Building a Second Brain</em> and the PARA framework. The vocabulary I disagree with on nested folders is still the vocabulary I’m disagreeing in.</li><li><strong>Brian Petro</strong> — Smart Connections, the Obsidian semantic-search plugin. The reason I felt comfortable skipping a separate vector DB was knowing this existed if I ever needed it inside Obsidian.</li><li><strong>Khoj.dev</strong> (YC W24) — the self-hosted-AI-second-brain option I considered and rejected. The fact that the rejection felt informed instead of defaulted is because their writeups are clear about what they do and don’t do.</li><li><strong>The Claude Code + Obsidian writers</strong> on Medium and Substack — the running thread that established this was a viable build at all. I’m one more voice in that conversation, not the first.</li></ul><h3>What this is the foundation for</h3><p>This post covers the architecture. The substrate. The decisions that determine what’s possible on top.</p><p>The operational layer — the eight skills I invoke from my phone or my desk to capture things, the trigger model that makes sure nothing falls through, the scheduled jobs that turn captured signal into Sunday-evening synthesis, the dashboard I look at every morning — is a separate post. Coming next.</p><p>Repo: <a href="https://github.com/devsandip/samaritan">github.com/devsandip/samaritan</a>. Plugin, contract, templates, the Cowork dashboard, and these blog posts as docs. Fork it, gut it, adapt it — that’s the point of shipping it open.</p><p>If you’ve built something in this space, especially a version of it that survived more than three months of real use, I’d love to hear what you kept and what you cut. The fail mode of a personal knowledge system is silent — you stop using it, you don’t notice for a while, the data calcifies, and then you’re back to ChatGPT amnesia. What kept yours alive?</p><p><em>Originally published at </em><a href="https://www.blog.sandip.dev/2026/05/30/second-brain-vector-db-skipped/"><em>https://www.blog.sandip.dev</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=32b69af7d752" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Scrask v4: An OpenClaw Skill That Turns Screenshots Into Calendar Events]]></title>
            <link>https://sandipdev.medium.com/scrask-v4-an-openclaw-skill-that-turns-screenshots-into-calendar-events-f21236ce823b?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/f21236ce823b</guid>
            <category><![CDATA[agent-platform]]></category>
            <category><![CDATA[openclaw]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Tue, 26 May 2026 04:39:39 GMT</pubDate>
            <atom:updated>2026-05-26T04:45:01.054Z</atom:updated>
            <content:encoded><![CDATA[<p>I screenshot conversations all the time. “Let’s meet next week.” “Send me your resume.” “Remind me next time we catch up.” Each one a small intention to remember something. Each one a screenshot that sits in my camera roll while the thing it was supposed to remind me about quietly passes.</p><p>So I built an OpenClaw skill to fix that. Called it Scrask. I feed it screenshots over Telegram most of the time, but the skill itself lives inside OpenClaw and works wherever OpenClaw does. v1 went up in February, a few weeks after OpenClaw launched and gave me the right primitives to do it cleanly. I expected a weekend toy. About a thousand people downloaded it — surprising enough to come back and give it a proper rewrite.</p><p>This post walks through what the skill does, what the original version looked like, and the three changes I’d most want a returning user to know about.</p><h3>What the skill does</h3><p>You give OpenClaw a screenshot — for me, that’s almost always forwarded to Telegram, but it works in any OpenClaw interface. A Python parser reads the image, pulls out the event or task — title, date, time, location, participants — and the calendar entry shows up on the other end. If something is ambiguous, the skill asks one targeted question; if not, it doesn’t.</p><p>The whole point is to skip the “now type all this in” step that sits between a plan being made in a group chat and a commitment landing on a calendar.</p><h3>What v3 actually did</h3><p>The original was a simple Python code. It took a screenshot, sent it to Claude vision (Gemini wasn’t in the picture yet), parsed an event or task out of the response, and wrote directly to Google Calendar or Google Tasks via the Google API. Configuration was an Anthropic API key plus a Google service-account JSON. The Telegram reply formatting lived inside the parser.</p><p>It worked. It was also brittle in three specific ways, and those three are what v4 addresses.</p><h3>v4.1 — Per-field confidence and targeted clarifications</h3><p>The parser used to return a single confidence score per item. So when a screenshot was ambiguous, the only move was to ask “is this right?” — which makes you re-read everything to figure out what it was actually unsure about.</p><p>v4.1 splits confidence per field. The parser now returns a confidences{} map for title, date, time, location, participants, description, and priority - plus two top-level scores: actionable_confidence (is this even an event or task?) and type_confidence (calendar event or task list item?).</p><p>Underneath the scores, each item carries a clarifications[] array of pre-formatted, targeted questions. The shape_intent function walks the mandatory fields for the inferred type and builds a question only for fields below threshold. If the title and date are confident but the time is shaky, you get &quot;What time is dinner with Priya?&quot;. If everything is confident, you get no question at all - most screenshots fall into this bucket.</p><h3>Why split perception from decisions</h3><p>Vision models are good at perception and bad at thresholds — ask one to apply a numeric cutoff and you’ll get a different answer depending on the weather. Code is good at thresholds and incapable of perception. Pulling fields out as data and applying cutoffs in Python means I can tune behaviour without re-prompting the model.</p><p>Three thresholds drive this — ACTIONABLE_THRESHOLD, TYPE_THRESHOLD, and FIELD_THRESHOLD - all configurable as CLI flags.</p><h3>v4.2 — Gemini is no longer required</h3><p>This is the most boring change and the one I think will matter most.</p><p>A new parse_with_openclaw() function reads OPENCLAW_VISION_PROVIDER, OPENCLAW_VISION_KEY, and an optional OPENCLAW_VISION_MODEL from environment variables OpenClaw injects. You can pin this mode explicitly with --provider openclaw.</p><p>If you bring your own Gemini or Claude key, auto-routing still applies — Gemini-first when both are present, Claude-only when Gemini misbehaves, OpenClaw’s default when neither is configured. The auto mode is now credential-aware: it routes based on what&#39;s in the environment instead of erroring out when a key is missing.</p><p>GEMINI_API_KEY moved from requires.env to optional_env. Lowering the floor for &quot;just try it&quot; mattered more than I expected.</p><h3>v4.3 — Hybrid invocation</h3><p>The original was a single-purpose script — you’d hand it a screenshot, it would run. Inside an agent platform like OpenClaw, that pattern doesn’t quite hold; the agent needs to know when to invoke Scrask versus another skill.</p><p>v4.3 adds a metadata.openclaw.invocation block to the skill manifest declaring mode: hybrid plus an alias list: scrask, scrask this, screenshot, screenshot to calendar. The platform contract is: check aliases first, fall back to implicit routing if no alias matched. The implicit conditions themselves are preserved verbatim, so any agent already routing on the old prose sees no behaviour change.</p><h3>The architectural shift: the parser writes nothing</h3><p>This is the change that doesn’t fit cleanly under any single version bump and matters more than any of them individually.</p><p>The v3 parser wrote directly to Google Calendar and Google Tasks. The v4 parser writes nothing. It emits a structured intent JSON — what the event is, where it should go, what confidence level it came in at — and OpenClaw delegates the actual save to whichever destination skill is installed: calctl for system calendars, things-mac for Things, apple-reminders, notion, whatever else you’ve wired up.</p><p>One screenshot in, one JSON plan out. Stateless. If you swap your task manager from Things to Notion next month, Scrask doesn’t change — the destination skill does. That’s the property I’m most pleased with, because it means I can ship the screenshot-reading half once and let the ecosystem handle the long tail.</p><h3>Documentation</h3><p>I also rewrote the docs from scratch, partly because the v3 docs no longer matched anything that was true and partly because I wanted a future me (or someone else picking this up) to be able to extend the parser without spelunking through commits.</p><p>The architecture overview is in two parts: part one is non-technical (what the skill does, where it lives, what each piece is for), part two is at the code level (file structure, the parser’s data contract, the shape_intent decision logic, why the parser is stateless). The decision flow is documented as a pair of Mermaid flowcharts - one for the parser, one for the bot side - plus a threshold reference table and three narrated end-to-end paths (happy, ambiguous, actionable-gate).</p><p>There’s also an interactive standalone HTML version. Mermaid v10 renders the flowcharts client-side, every decision node and threshold-table row is clickable, and the popups explain what each gate does with references back into the code. It’s self-contained — open the HTML file in a browser and it works.</p><h3>Try it. Tell me what breaks.</h3><p>The repo is at <a href="https://github.com/devsandip/scrask-bot">github.com/devsandip/scrask-bot</a>.</p><p>Skill in Claw Hub: <a href="https://clawhub.ai/devsandip/scrask-bot">https://clawhub.ai/devsandip/scrask-bot</a></p><p>The most useful thing you can do is run it on a screenshot you’d actually want on your calendar and tell me what doesn’t work. I’m especially interested in the screenshots I haven’t tested well — non-English text, dense flyers, low-contrast captures, half-cropped invites, the Telegram formatting where the date is in one bubble and the time is in the next. Edge cases reported now are worth far more than edge cases discovered quietly six months from now.</p><p><em>Originally published at </em><a href="https://www.blog.sandip.dev/2026/05/26/scrask-v4-openclaw-skill-screenshots-to-calendar/"><em>https://www.blog.sandip.dev</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f21236ce823b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A PM Building Solo: Three Weeks of No-Op Deploys and the Discipline I Skipped]]></title>
            <link>https://sandipdev.medium.com/a-pm-building-solo-three-weeks-of-no-op-deploys-and-the-discipline-i-skipped-417987a412d9?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/417987a412d9</guid>
            <category><![CDATA[solobuilder]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[claude-code]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Sun, 17 May 2026 17:26:40 GMT</pubDate>
            <atom:updated>2026-05-17T17:26:40.636Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*kNat558SxI_oxarh" /></figure><p>I have shipped products at Amazon for nine years. I sat in standups, signed off on PRs, watched engineers argue about deployment safety, read RCAs after outages. I thought I understood the discipline.</p><p>Today I spent the whole day undoing three weeks of consequences that proved I did not.</p><p>This is the writeup of one day untangling the messes a solo founder makes when they have the intellectual knowledge of engineering discipline but none of the institutional guardrails that enforce it. Every single one of these mistakes would have been caught in five minutes by a real team. None of them were caught for weeks, because the team is one person, and the one person was me.</p><h3>The Day the Deploy Script Lied</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vfNZADTWhjuIcF82" /></figure><p>Between April 24 and May 17, I shipped backend changes to my side project most days. My deploy-backend.sh reported success every time. It even had a verification step at the end.</p><p>This morning I realised the verification step was lying.</p><p>The script was pushing a new Docker image to ECR and calling update-service --force-new-deployment, which rolls the running task. But it never updated the task-definition&#39;s pinned image. The task kept rolling onto the same April 24 image. Three weeks of code changes never reached a user.</p><p>In a real eng org, somebody would have noticed. A QA pass. A second engineer asking, “wait, did the SHA change?” A customer raising a bug that should have been fixed last week. Solo, the only check is whether I personally remembered to run aws ecs describe-services after each deploy. I didn&#39;t. Not once in 23 days.</p><p>Fixed with two PRs. The script now SHA-pins the image, registers a new task-definition revision, and verifies the running task is on that revision before exiting. The verification is what makes it safe. Before today, “verification” meant “the script exited 0.”</p><h3>The Branches That Quietly Drifted</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*zDPOYNCRq0742XaZ" /></figure><p>I run a next branch (Amplify auto-deploys it to my staging subdomain) and a main branch (prod). Promotion is supposed to be a next to main PR every single time.</p><p>A few PRs ago I got impatient with a small docs change and merged direct to main. Then again with another bundle. Then once more. None of them got backported to next.</p><p>By today, the branches differed on seven files. Six existed only on main. That meant my next planned promotion would have <em>deleted</em> those six files from prod, because the next branch genuinely did not know they existed.</p><p>I caught it because I happened to diff the two branches before merging. Nothing structural would have stopped me from shipping the regression. GitHub branch protection requires a paid Team plan on private repos. I don’t have one, so gh pr merge --admin happily merges PRs with red or pending checks. There is no enforcement.</p><h3>The Doc That Hallucinated Its Own Reality</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*BqzTsp6lrI6_dBHD" /></figure><p>The repo has a CLAUDE.md file, which is the doc my AI coding assistant reads to understand the project. It is supposed to be the canonical answer to &quot;where are we?&quot;</p><p>When I read it cold this morning, it was wrong in eight separate ways.</p><ul><li>It pointed at a worktree path that no longer existed</li><li>It pointed at the wrong path for a publisher CLI</li><li>A stale V2 commit pointer 19 commits behind reality</li><li>A dead localhost:3001 ritual nobody ran any more</li><li>Internal contradiction inside the same section about whether Phase 2 was done</li><li>V2 still presented as in-flight, weeks after it shipped</li><li>Em-dashes and emojis the folder-level rule explicitly forbids</li></ul><p>The cost wasn’t reading time. The cost was that every Claude Code session for the past month had been building plans on hallucinated context. I was getting confident answers rooted in a version of the repo that had not existed for over a month.</p><p>Rewrote it end to end. 281 lines. +226/-156. Now reflects reality. The reason it rotted: nobody read it cold to check. There’s only me, and I am the one whose work was making it stale.</p><h3>The Files That Existed Only in Finder</h3><p>I had deleted two HTML mockup files locally weeks ago because they were V2 review artifacts and the review was done. I deleted them in Finder, not in git. They sat as uncommitted deletions in my working tree until today, when a routine git pull --rebase blew up because of them.</p><p>The fix took thirty seconds. The lesson is bigger: any manual operation on a tracked directory creates a future failure mode. Always. Use git rm or do not touch the file.</p><h3>The .gitignore That Did Not Ignore</h3><p>I started using a RESUME.md file at the repo root as a session-handoff document between Claude Code sessions. Personal session state, not project truth. I added it to .gitignore today, expecting that to be enough.</p><p>It was not. The file had been committed back in April, and .gitignore only blocks <em>new</em> tracking. Already-tracked files stay tracked. I had to ship a separate PR with git rm --cached RESUME.md for the rule to take effect. Then I had to manually unblock my own pull on the main clone because the incoming change was about to delete the file I was actively using.</p><p>Forty minutes of yak-shaving for a one-line gitignore that should have existed since April.</p><h3>What This Taught a PM</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*nB0MVXje7nIHlHH9" /></figure><p>Every one of these messes was preventable with discipline I have known about for over a decade. PRs need second eyes. Direct-to-main merges need same-day backports. Docs go stale and need periodic audits. Tooling needs verification, not just optimism. git rm, not Finder.</p><p>So why didn’t I do them? Because solo means there is no system enforcing them. In an eng org, the guardrails are people. The reviewer who pushes back. The QA pass. The on-call who pages you when prod drifts. The teammate who diffs your PR before you merge.</p><p>Solo, the guardrails are nothing unless you build them as code or rituals you actually follow.</p><p>The intellectual understanding and the muscle memory are not the same skill. As a PM I could have written a Confluence page on every one of these practices. As an engineer of one, I trusted velocity and skipped them.</p><h3>What I Shipped Today</h3><p>14 PRs of hardening across two sessions. The shape of it:</p><ul><li>A GitHub Actions cron that diffs main and next weekly and opens a tracking issue when they diverge</li><li>A safe-merge.sh wrapper that warns and sleeps 10 seconds before any non-next merge to main</li><li>A new CLAUDE.md rule that direct-to-main merges must be backported same-day</li><li>A SHA-pinned deploy script with mandatory post-deploy verification</li><li>A rewritten CLAUDE.md that reflects reality</li><li>A documented session-resume convention with proper .gitignore anchoring</li><li>The cleanup of every regression listed above</li></ul><h3>The Lesson, Plainly</h3><p>For PMs thinking about building solo, especially in the AI-coding era when the velocity feels infinite: you will rebuild the entire institutional guardrail stack of your old company, or you will eat the cost of not having it.</p><p>Today was my bill. It came due all at once, on a day where I shipped zero product features. The hardening I shipped is the discipline I should have had on day one. The day to build it was always day one, and there is no point pretending otherwise.</p><p><em>Originally published at </em><a href="https://www.blog.sandip.dev/2026/05/17/pm-solo-builder-no-op-deploys/"><em>https://www.blog.sandip.dev</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=417987a412d9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I Built a Private “AI News Radio” for $0/Month]]></title>
            <link>https://sandipdev.medium.com/tutorial-how-i-built-a-private-ai-news-radio-for-0-month-7d0fa146acd1?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/7d0fa146acd1</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[llm-applications]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Thu, 29 Jan 2026 17:58:12 GMT</pubDate>
            <atom:updated>2026-01-29T22:42:35.664Z</atom:updated>
            <content:encoded><![CDATA[<p>I have a weird morning routine. I read general news at 5:30 AM, tech news at 11:00 AM, and startup news at 1:00 PM. But clicking through 15 different newsletters was killing my productivity.</p><p>I wanted a “Magic Radio” that would:</p><ol><li><strong>Read my emails for me.</strong></li><li><strong>Filter them</strong> into “General,” “Tech,” and “Startup” buckets.</li><li><strong>Turn them into a podcast</strong> that I can listen to while working out or commuting.</li><li><strong>Hide the Tech news</strong> until 11:00 AM so I don’t get distracted.</li></ol><p>No app existed that could do this. <strong>So I built it myself without writing a single line of code. All in 2 hours.</strong></p><p>Here is exactly how you can build this “AI News Router” using <strong>Make.com</strong>, <strong>OpenAI</strong>, and <strong>JustCast</strong>.</p><h3>The Stack</h3><ul><li><strong>Make.com:</strong> The “Glue” that connects everything (Free tier: 1,000 ops/month).</li><li><strong>OpenAI API:</strong> The “Brain” (gpt-4o-mini) and “Voice” (TTS). Cost: ~$0.50/month.</li><li><strong>Dropbox:</strong> The “Storage” (Free).</li><li><strong>JustCast:</strong> The “Broadcaster” that turns Dropbox folders into Podcast feeds (Free).</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ut54tNp_JfM6QS5T-bpydA.png" /></figure><h3>Phase 1: The Setup (5 Minutes)</h3><h4>1. Prepare Gmail</h4><ul><li>Create a specific label in Gmail (e.g., Newsletters).</li><li>Set up filters in Gmail to automatically apply this label to your favorite newsletters (e.g., <em>Morning Brew</em>, <em>Stratechery</em>, <em>The Verge</em>).</li><li><strong>Pro Tip:</strong> Mark them as “Skip Inbox” so they don’t clutter your actual inbox.</li></ul><h4>2. Prepare Dropbox</h4><ul><li>Create a folder named AI_Radio.</li><li>Inside it, create sub-folders for your buckets: /Morning, /Tech, /Startup, and /Opinion.</li></ul><h4>3. Prepare JustCast</h4><ul><li>Go to <a href="https://justcast.com/">JustCast.com</a> and connect it to your Dropbox.</li><li>It will automatically detect those folders and give you an <strong>RSS Feed URL</strong> for each one.</li><li><em>Note:</em> The free tier limits you to the <strong>3 most recent items</strong>, which is perfect for news (you don’t want yesterday’s news anyway).</li></ul><h3>Phase 2: The Logic (Make.com)</h3><p>Create a new Scenario in Make.com and follow these modules.</p><h4>Step 1: The Trigger (Gmail)</h4><ul><li><strong>Module:</strong> Gmail &gt; Watch Emails</li><li><strong>Folder:</strong> Select your Newsletters label.</li><li><strong>Criteria:</strong> “Only Unread” (to avoid re-processing old stuff).</li><li><strong>Max Results:</strong> Set to 20 (Safety buffer for busy days).</li></ul><h4>Step 2: The Intelligence (OpenAI)</h4><ul><li><strong>Module:</strong> OpenAI &gt; Create a Completion (Chat)</li><li><strong>Model:</strong> gpt-4o-mini (It’s fast, cheap, and smart).</li><li><strong>Response Format:</strong> Set to <strong>JSON Object</strong> (Crucial!).</li><li><strong>System Prompt:</strong></li></ul><blockquote>You are a newsletter parser. Output a valid JSON object with 3 keys:</blockquote><blockquote>“category”: Strictly choose ‘Morning’, ‘Tech’, or ‘Startup’.</blockquote><blockquote>“news_script”: Extract hard news/facts. Rewrite as a news anchor script. Remove URLs.</blockquote><blockquote>“opinion_script”: Extract editorial/analysis. Rewrite as a podcast host script.</blockquote><ul><li><strong>User Prompt:</strong> Drag and drop the Text Body variable from the Gmail module.</li></ul><h4>Step 3: The Parser</h4><ul><li><strong>Module:</strong> JSON &gt; Parse JSON</li><li><strong>Input:</strong> Drag the Content output from the OpenAI module.</li><li><em>Why:</em> This splits the AI’s answer into usable variables (category, news_script, etc.).</li></ul><h4>Step 4: The Router</h4><ul><li><strong>Module:</strong> Flow Control &gt; Router</li><li>Connect 4 paths to this router.</li></ul><h3>Phase 3: The Routing Logic</h3><p>Now we tell the bot where to send the audio.</p><h4>Path A: Morning News</h4><ul><li><strong>Filter:</strong> category = Morning AND news_script exists.</li><li><strong>Module 1:</strong> OpenAI &gt; Create Speech<br>-<strong>Input:</strong> news_script<br>-<strong>Voice:</strong> alloy (Crisp, newsy voice).</li><li><strong>Module 2:</strong> Dropbox &gt; Upload File<br>-<strong>Folder:</strong> /AI_Radio/Morning<br>-<strong>Filename:</strong> {{now}}-Morning.mp3 (Using a timestamp prevents overwriting).</li></ul><h4>Path B: Tech News</h4><ul><li><strong>Filter:</strong> category = Tech AND news_script exists.</li><li><strong>Module 1:</strong> OpenAI &gt; Create Speech<br>-<strong>Input:</strong> news_script<br>-<strong>Voice:</strong> echo (Deeper, tech-focused voice).</li><li><strong>Module 2:</strong> Dropbox &gt; Upload File<br>-<strong>Folder:</strong> /AI_Radio/Tech</li></ul><h4>Path C: The “Opinion” Catch-All</h4><ul><li><strong>Filter:</strong> opinion_script exists.</li><li><strong>Module 1:</strong> OpenAI &gt; Create Speech<br>-<strong>Input:</strong> opinion_script<br>-<strong>Voice:</strong> onyx (Storyteller voice).</li><li><strong>Module 2:</strong> Dropbox &gt; Upload File<br>-<strong>Folder:</strong> /AI_Radio/Opinion</li></ul><h3>Phase 4: The “Set and Forget”</h3><ol><li><strong>Schedule:</strong> Set the Make.com scenario to run <strong>Every Day at 4:30 AM</strong>.</li><li><strong>Subscribe:</strong> Open your podcast app (Apple Podcasts, Pocket Casts, etc.) and add the RSS URLs from JustCast.</li></ol><p><strong>The Result:</strong></p><p>When I wake up at 5:30 AM, my “Morning News” feed has a fresh 10-minute episode summarizing world events. At 11:00 AM, my “Tech” feed updates with the latest from Silicon Valley.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*28s7Z0KW4xgWywaMexm57Q.png" /><figcaption>Make.com scenario</figcaption></figure><h3>Total Cost Breakdown</h3><ul><li><strong>Make.com:</strong> Free (The 1,000 ops cover ~30 newsletters/day easily).</li><li><strong>Dropbox:</strong> Free (2GB is plenty for MP3s).</li><li><strong>JustCast:</strong> Free.</li><li><strong>OpenAI:</strong> ~$0.50/month (You only pay for the text you generate audio for).</li></ul><p><strong>Why this is better than an App:</strong></p><p>Apps try to guess what you want. This system does exactly what <strong>you</strong> tell it to do. It respects your time, your buckets, and your routine.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2YN5-IpfQVtPRGmyE9r8Pw.png" /><figcaption>And here is my podcast</figcaption></figure><h3>Future Roadmap (If I Decide to Over-Optimize This Further)</h3><p>There are a few upgrades I might add if I feel like pushing this into truly unreasonable territory.</p><p>One is a “bullet brief” mode: a 30-second headline rundown for mornings when even a full audio briefing feels ambitious. Another is basic de-duplication, so the same story repeated across six newsletters gets compressed instead of re-read. I’m also tempted to add confidence scoring, so only news the model is reasonably sure about gets auto-published. And finally, I’d like to tweak the opinion prompt further to slow the pacing and soften the tone for late-night listening.</p><p>None of this is necessary. The system already does what I wanted it to do.</p><h3>Links</h3><ol><li>Make.com scenario: <a href="https://us2.make.com/public/shared-scenario/AIxcSbu7hBk/my-newsletter-podcast">https://us2.make.com/public/shared-scenario/AIxcSbu7hBk/my-newsletter-podcast</a></li><li>My podcast<br>- Morning — <a href="https://podcasts.justcast.com/morning">https://podcasts.justcast.com/morning</a><br>- Opinion — <a href="https://podcasts.justcast.com/opinion">https://podcasts.justcast.com/opinion</a><br>- Startups — <a href="https://podcasts.justcast.com/startup">https://podcasts.justcast.com/startup</a><br>- Tech — <a href="https://podcasts.justcast.com/tech">https://podcasts.justcast.com/tech</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7d0fa146acd1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Curry in the wild: Parathas under the Pines]]></title>
            <link>https://sandipdev.medium.com/curry-in-the-wild-parathas-under-the-pines-d383d31890ac?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/d383d31890ac</guid>
            <category><![CDATA[indian-food]]></category>
            <category><![CDATA[camping]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Sun, 05 Oct 2025 18:03:04 GMT</pubDate>
            <atom:updated>2025-10-05T18:15:47.842Z</atom:updated>
            <content:encoded><![CDATA[<p>There are few spectacles in this age of mechanical excess so instructive as the modern desi <a href="https://www.msn.com/en-in/news/world/pakistani-canadian-familys-over-the-top-camping-feast-in-middle-of-forest-catches-internets-eye-video/ar-AA1NECeF">family gone camping</a>. They arrive not as pilgrims into the wilderness, but as conquerors bearing condiments. The trees stand mute; the rivers, unconsulted. And from the tailgate of a vehicle the colour of prosperity emerges a glittering procession of civilisation in miniature — gas burners, pressure cookers, chopping boards, crates of tomatoes, and that inevitable steel masala box, polished to a missionary’s zeal.</p><p>In a clearing meant for tents, <a href="https://youtu.be/EQ-Xr1DR46Y?si=JF3BzDRG_ftrdsZb">a full kitchen is raised like a fort</a>. The folding tables are spread; the propane tanks hiss like dragons at dawn; onions fall beneath the knife with military precision. The air, which once carried only pine and rain, now bears the heady odour of cumin and ghee — the incense of man’s refusal to be humbled by creation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hMr0MDNGHJ7m03uoUEfXwg.png" /></figure><h3>I. The Fear of Simplicity</h3><p>Camping was once a kind of fast. One left behind the furniture of life to rediscover its floorboards. But the modern Indian, terrified of deprivation, must bring the dining room along, complete with curtains of smoke and the ceaseless clatter of spoons. He has mistaken inconvenience for tragedy. The absence of a second curry is felt as a moral outrage.</p><p>He speaks, with wounded pride, of “roughing it,” yet roughs nothing. His tents have carpets, his chairs have arm-rests, and his heart — alas — has no room for wonder.</p><h3>II. The Domestic Empire</h3><p>Observe the ritual as it unfolds. The men open beer cans with the air of philosophers beginning a debate on leisure. One wanders away on a “hike,” which is to say, a brief constitutional designed to return in time for the next course. Another reclines in a camp chair like a sultan exiled to suburbia. Their kingdom prospers without effort, for the real labourers are close at hand.</p><p>Beneath a tree, <a href="https://youtube.com/shorts/0DJJia4-zQY?si=wf46ExFHZlXoXI_A">a woman crouches before a portable stove</a>, her hair escaping in the wind like a flag of quiet rebellion.<a href="https://youtube.com/shorts/SrBvDlyFdh8?si=A_xX6d2aMnkV0W3d"> Another rinses pans</a> with bottled water, cursing softly as the <a href="https://youtube.com/shorts/ylupCuLg9co?si=fb-49IpjPl3owtXz">grease clings more faithfully than her husband’s gratitude</a>. Their laughter, when it comes, is the laughter of habit — cheerful servitude rehearsed so long it feels like joy.</p><p>Thus the empire of patriarchy survives the voyage better than the crockery; it emerges unbroken from every suitcase.</p><h3>III. The Great Refusal to Finish</h3><p>When the feast is done and the men grow eloquent with beer, the wilderness stands transformed into a battlefield of plastic plates, oily ladles, and rice that clings like guilt to every surface. The fire dies, but the pots remain — blackened relics of comfort’s campaign.</p><p><a href="https://youtube.com/shorts/-pHt5puhX-s?si=BDCpUybnfM59MuoB">Cleanup, like justice, is deferred indefinitely.</a> A token effort is made with tissues and bottled water until fatigue — or apathy — declares victory. The detritus of civilisation is left as tribute to the forest, which accepts it silently, as it once accepted ashes and bones.</p><p>Man has learned to conquer nature, <a href="https://www.msn.com/en-in/news/India/outrage-as-tourists-litter-indias-scenic-mountains-with-gutka-stains-plastic-waste/ar-AA1BKrlb?apiversion=v2&amp;noservercache=1&amp;domshim=1&amp;renderwebcomponents=1&amp;wcseo=1&amp;batchservertelemetry=1&amp;noservertelemetry=1">but not to tidy after himself.</a></p><h3>IV. The Philosophy Mislaid</h3><p>India once taught the world the virtue of restraint. The saint in the forest ate nothing, owned nothing, feared nothing. His reward was revelation. His descendants, returning to the same forest, bring fifteen kilograms of marinated meat and a Bluetooth speaker. They call it progress.</p><p>The Western man goes into the wild to discover how little he needs. The Indian goes to prove how much he can manage to carry. He builds a temple of aluminium foil and Tupperware, then prays to it thrice daily.</p><h3>V. The True Wilderness</h3><p>I have no quarrel with food; I quarrel with the fear that lies beneath it. For this elaborate feast is not hospitality but defence — an edible barricade against the terror of simplicity. We cannot sit quietly before the mountains, lest we hear ourselves think. We cannot gaze at the stars without demanding dessert.</p><p>The wilderness asks of us only humility; we answer with recipes.</p><h3>VI. Conclusion</h3><p>I say nothing against the curry itself, which is admirable in all seasons and latitudes. But I say this: a man who must carry his kitchen into the forest has mistaken comfort for culture and appetite for achievement. His wife, sweating before the stove while he drinks beneath the stars, is not enjoying leisure; she is furnishing his illusion of it.</p><p>When future archaeologists unearth our age, they will find in the ashes of our campfires not bones or arrows, but half-melted spoons and unopened packets of spice. And they will know that we loved civilisation too much to ever leave it — even for a weekend.</p><p>For the true wilderness is not the place without food; it is the place without vanity. And we, alas, cannot seem to find it.</p><p>#CulturalCommentary #IndianDiaspora #CampingCulture #GenderRoles</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d383d31890ac" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Thiel Delusion: Why Deep Research Still Builds the Future]]></title>
            <link>https://sandipdev.medium.com/the-thiel-delusion-why-deep-research-still-builds-the-future-4fdaa4a1001e?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/4fdaa4a1001e</guid>
            <category><![CDATA[higher-education]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Tue, 19 Aug 2025 16:56:31 GMT</pubDate>
            <atom:updated>2025-08-19T16:56:31.648Z</atom:updated>
            <content:encoded><![CDATA[<p>For almost 2 decades now, Peter Thiel has been crusading against colleges and higher education, calling it a scam no less. His eponymous Peter Thiel scholarship encouraged smart kids to forego higher education to found startups that largely ended up being middling at best. Only 11 out of the 290 fellows founded startups that reached unicorn ( $1 Billion +) valuation — notable among them being Ethereum, Figma and Scale AI. Critics of Thiel’s fellowship have argued, among other things, that transformative change, the kind that expands the horizon of what is possible and moves the needle for humanity as a whole, can only come from Deep Research. The next big thing would come from a lab, not a co-working space.</p><p>And the best example of that is the Thiel funded startup, OpenAI. Look at their top people, everyone (with the notable exception of the CEO, Sam Altman) has a PhD and years of research experience. Meta was recently in the news for poaching execs from OpenAI and other AI startups with packages of over $100 million — again almost all PhDs.</p><p>In other words, Thiel’s own investments tell a different story than his ideology. OpenAI, perhaps the most consequential company of this decade — and one Thiel helped fund — wasn’t built by dropouts tinkering in a garage. Now yes, Sam Altman, CEO of OpenAI, is himself a college dropout. But let’s be clear, he’s not the guy building models. He’s the guy hiring the people who spent a decade doing the science. Altman dropped out of Stanford but surrounded himself with PhDs, researchers, and ex-academics who actually push the frontier. His success isn’t an argument against higher education; it’s a case study in knowing who to bet on.</p><p>If there’s a lesson in all this, and I am told all Linkedin posts need to conclude with learnings, it’s that shortcut thinking rarely leads to long-term impact. Chasing valuations might get you a headline, but changing the world requires depth, discipline, and a stubborn commitment to the long haul. The most transformative technologies of our time -mRNA vaccines, CRISPR, LLMs -didn’t come from dorm rooms or co-working spaces. These innovations weren’t built by people who did a 4 week crash course on programming. They came from people who stayed the course, put in the years, and did the hard, often thankless work of real research. So no, kids (or parents of kids if you are listening, although high school kids are increasingly creating LinkedIn profiles apparently) shouldn’t drop out. If you truly want to build the future, go get the education, grind in the lab, and stop looking for the fast lane to billion dollar exits.</p><p>#HigherEducation #ThielFellowship #Startups #DeepResearch #OpenAI #LongTermThinking #AcademicInnovation #FutureOfWork #ArtificialIntelligence</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4fdaa4a1001e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Big Basket, hum aapke hai kaun?]]></title>
            <link>https://sandipdev.medium.com/big-basket-hum-aapke-hai-kaun-fa69cd79e9de?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/fa69cd79e9de</guid>
            <category><![CDATA[customer-experience]]></category>
            <category><![CDATA[online-retail]]></category>
            <category><![CDATA[retail]]></category>
            <category><![CDATA[bigbasket]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Thu, 19 Jan 2017 14:20:50 GMT</pubDate>
            <atom:updated>2017-01-20T05:50:29.814Z</atom:updated>
            <content:encoded><![CDATA[<p>Big Basket, hum aapke hai kaun?</p><p>There is probably no app I use more than BigBasket. Yes, you are probably thinking “Single guy, does he ever cook? What does he shop from BigBasket? Coke?”. And that was true until 6 months. Now I do order vegetables and stuff from BigB, a lot of them. It used to be Swiggy until 6 months ago but now Swiggy has been replaced by BigBasket.</p><p>And that’s when it’s flaws have become evident and by far all of them stem from their inability/refusal to listen to the customer.</p><h4>Warehousing issues</h4><p>My two biggest problems with them recently have been — items covered in dust ostensibly because they gathered dust in their warehouse and secondly delivering product that are near, if not past, their expiry date. I have received products that are 18 months old and BigBasket has been kind enough to take a return, no question asked. Good BigB. I have provided feedback to them a couple of times on both the issues — old products and products covered with dust. But it still keeps happening. Bad BigB. I don’t know how they will solve these problems but I sure don’t like to wipe all my products with cloth after every purchase while also checking their expiry date</p><h4>Data ignorance</h4><p>Now I have been a BigB customer for over 3 years and they know how often and what I buy, which brand, what quantity. They know, or ought to know, how long a 5 kg packet of rice or a 200ml bottle of shampoo lasts in my house. And yet when I place a order on their site, their suggestions are either the most common items that I buy or items from my last order. Now BigB, why would I need to order 2 kilos of potato again in 5 days? This is the age of data. Companies are using data to gain every advantage against their competitors and yet you just let all that data sit idle? That’s a crying shame.</p><p>If you used that data properly, you would just know how long each items lasts in my house (5kg rice every 3 months, potatoes every 3–4 weeks, body wash lasts 2 months etc) but a lot lot more. Just look at retails stores like Target and what they are doing with data. For instance, if you see I have started ordering more green veggies and less potatoes, you could figure out that I am dieting. Now with that information, you could suggest me related stuff like brocolli, supplements and even hand sanitizer, something people often use at the gym.</p><p>And of course you can use that data to get ad revenue. Heck, you can help your partner brands to send new free product samples to exactly the right customer by diving into your customer data (and with due regard to their privacy obviously). Because of the nature of your business, you have far granular and accurate data than any brick and mortar retail store in the country. Make use of it.</p><h4>Know your customer</h4><p>A couple of months ago BigB delivered me a bag of rice that had insects in it. Because there were insects, I threw the bag and asked for a refund or replacement. BigBasket initially said they can’t provide a refund/replacement because I threw away the bag. While I completely understand their concern here, I felt a bit insulted because some guy in BigBasket thinks that I, a customer who has been buying exclusively from them (I rarely go to retail stores) for over 3 years, is lying to score a Rs 400 bag of rice. They did refund me after I reminded them how long I have been a customer and I am deeply grateful for the prompt action thereafter. But did I really have to go through all that? Wouldn’t a quick check at my account on their end revealed that this guy is very likely not lying?</p><p>And then during demonetization, my orders were either delayed, had missing items (which BigBasket had informed me a couple of hours before delivery) or worst of all, my delivery slot was changed without informing me. Forsaking your old customers for new ones, who will probably go back to buying from kirana stores soon, is a double edged sword. Sure you get new customers but you lose the trust of the old ones and they would be happy to switch once another competitor comes along. I know I will, provided the customer service is better.</p><p>Despite all that, their customer service is definitely prompt. I will give them that</p><h4>Templatized Social Media presence</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*yD04bbR7jMnaMbmgOZN0Iw.jpeg" /></figure><p>How can I give an order ID for an order that I can’t place because the item is out of stock? I rest my case.</p><h4>Prognosis</h4><p>BigBasket is a new age company but seems to thinks slightly (not entirely) like the retail stores of yore. They have great discounts, stock the right products all the time (except whipping cream and bacon which are usually unavailable) and are frugal but don’t listen to customer feedback. Its not that they don’t understand customer needs, they do. But they don’t go much beyond it. The buzzword today is “customer delight”, and while I have been always satisfied with BigBasket, I have rarely ever been delighted.</p><p>Some of it probably has to do with their leadership structure- they are largely veterans of the retail and FMCG world. These sectors don’t require responsive customer service (they do obviously require understanding the customer which BigB does well). These sectors also don’t require data science to the extent Silicon Valley and Indian tech startups use. And a LinkedIn search (albeit imperfect) tells me they have less than a handful of people doing Data Science for them. It might do them well to look into the ethos of customer responsive companies of the digital age. Customers today expect a lot more and they are very vocal about any displeasure. Heck some of these pesky customers even end up writing a huge ass post on Medium.</p><p>Here is the thing BigBasket, you have the right business model, unlike Grofers which buys from retailstores thereby foregoing the massive discounts that wholesalers and distributors get while also assuming the transportation cost of delivering product to the customer, often in smaller batches. BigBasket can cut out, not just the retailer but also the wholesaler on one end and also bundle multiple orders to save costs at the other end, because it invested in a warehouse. Now you just need to get to be more responsive to customers.</p><p>So BigB, hum aapke hai kaun? Are we in a serious long term relationship or are we into a short term chicken &amp; ham thank you for the groceries ma’am kinda relationship?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fa69cd79e9de" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Uber made me realize I need a car]]></title>
            <link>https://sandipdev.medium.com/uber-made-me-realize-i-need-a-car-a4130f57446?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/a4130f57446</guid>
            <category><![CDATA[uber]]></category>
            <category><![CDATA[olacabs]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Fri, 07 Oct 2016 12:00:18 GMT</pubDate>
            <atom:updated>2016-10-07T12:19:50.011Z</atom:updated>
            <content:encoded><![CDATA[<p>Thanks to Uber I have decided to buy a car</p><p>Yes you heard that right. I am planning to finally get myself a car. No, I am not planning on becoming a Uber “driver partner” (a clever way for Uber to tell you they don’t own any responsibility for the driver’s conduct). I am tired, annoyed and frustrated with the Uber experience. Here’s what I have to face on a near daily basis</p><ol><li>Not a single driver knows how to use the GPS navigation on their phone. So I have to explain my address and guide them every time, every single bloody fucking time</li><li>They have put extremely novice drivers on the road. They drive erratically, often reckless and they don’t know Bangalore landmarks. Once a driver asked me where Brigade Road was. When Uber is sending Uber Go drivers, they are not sending their best. They are sending driver that are novice, arrogant and some I assume are good drivers. [See what i did there. Just channeled my inner Trump]</li><li>The drivers will greedily accept a trip and then they won’t come for pick up and won’t cancel either. And if you cancel, Uber will charge you hefty cancellation fee. This is more the case with Uber share which my girlfriend uses to go to work everyday. Threatening the driver by saying that you will complain also doesn’t work because they know Uber won’t take action. This is a problem borne out of User’s dumb incentive structure but Uber doesn’t give a shit because right now, it’s busy trying to outdo Ola.</li><li>Language barrier. Yes, I know I can’t expect every Bangalore driver to know Hindi or English. But some Uber drivers don’t even know Kannada, they only and only speak Tamil. I am a Bengali who is from Assam; I already speak 4 Indian languages. How many more do I need to learn?</li><li>Dumb routing algorithm. I took a cab from my place in Begur to my office in Brigade Gateway Rajajinagar. While going Uber cab travelled a distance of about 25km but while coming back it did a distance of 50km. That’s because Uber app decided to take the NICE road (which has tolls of which I wasn’t informed) and it’s a near 25km detour. The driver also hadn’t realised it and the cab was nearly out of fuel halfway and we had to switch off the AC, pull up the windows and travel a large part of the distance in neutral to reach our destination. All because Uber ‘s algo decided to sort routes by duration instead of distance. Also the routing algorithm that directs the driver has issues. I have noticed that the suggested routes in many cases had completely unnecessary detours and in a few cases made the driver go full circle and return at the same place.</li></ol><p>I bought a motorcycle 4 years ago because I hated haggling with Bangalore auto drivers. Now, I am planning to buy a car because I hate the Uber experience</p><p>PS: Ola, though not a lot better, still seems to have somewhat better trained drivers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a4130f57446" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Thanks to Uber I have decided to buy a car]]></title>
            <link>https://sandipdev.medium.com/thanks-to-uber-i-have-decided-to-buy-a-car-e0130b54b398?source=rss-9cf403e37e8e------2</link>
            <guid isPermaLink="false">https://medium.com/p/e0130b54b398</guid>
            <category><![CDATA[olacabs]]></category>
            <category><![CDATA[taxi]]></category>
            <category><![CDATA[uber]]></category>
            <dc:creator><![CDATA[Sandip Dev]]></dc:creator>
            <pubDate>Fri, 07 Oct 2016 04:54:49 GMT</pubDate>
            <atom:updated>2016-10-07T04:54:49.479Z</atom:updated>
            <content:encoded><![CDATA[<p>Thanks to Uber I have decided to buy a car</p><p>Yes you heard that right. I am planning to finally get myself a car. No, I am not planning on becoming a Uber “driver partner” (a clever way for Uber to tell you they don’t own an responsibility for the driver’s conduct). I am tired, annoyed and frustrated with the Uber experience. Here’s what I have to face on a near daily basis</p><ol><li>Not a single driver knows how to use the GPS navigation on their phone. So I have to explain my address and guide them every time, every single bloody fucking time</li><li>They have put extremely novice drivers on the road. They drive erratically, often reckless and they don’t know Bangalore landmarks. Once a driver asked me where Brigade Road was. When Uber is sending Uber Go drivers, they are not sending their best. They are sending driver that are novice, arrogant and some I assume are good drivers. [See what i did there. Just channeled my inner Trump]</li><li>The drivers will greedily accept a trip and then they won’t come for pick up and won’t cancel either. And if you cancel, Uber will charge you hefty cancellation fee. This is more the case with Uber share which my girlfriend uses to go to work everyday. Threatening the driver by saying that you will complain also doesn’t work because they know Uber won’t take action. This is a problem borne out of User’s dumb incentive structure but Uber doesn’t give a shit because right now, it’s busy trying to outdo Ola.</li><li>Language barrier. Yes, I know I can’t expect every Bangalore driver to know Hindi or English. But some Uber drivers don’t even know Kannada, they only and only speak Tamil. I am a Bengali who is from Assam; I already speak 4 Indian languages. How many more do I need to learn?</li><li>Dumb routing algorithm. I took a cab from my place in Begur to my office in Brigade Gateway Rajajinagar. While going Uber cab travelled a distance of about 25km but while coming back it did a distance of 50km. That’s because Uber app decided to take the NICE road (which has tolls of which I wasn’t informed) and it’s a near 25km detour. The driver also hadn’t realised it and the cab was nearly out of fuel halfway and we had to switch off the AC, pull up the windows and travel a large part of the distance in neutral to read our destination. All because Uber &#39;s algo decide to sort routes by duration instead of distance. Also the routing algorithm that directs the driver has issues. I have noticed that the suggested routes in many cases had completely unnecessary detours and in a few cases made the driver go full circle and return at the same place.</li></ol><p>I bought a motorcycle 4 years ago because I hated haggling with Bangalore auto drivers. Now, I am planning to buy a car because I hate the Uber experience</p><p>PS: Ola, though not a lot better, still seems to have somewhat better trained drivers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e0130b54b398" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>