Knowledge Jam, March 20, 2026

Building software
just for me in 2026

Ashish Acharya

"I wish there was an app for this."

I love building software to solve niche problems.
With AI, I can now do this at scale.


Today I'll show you five things I built.

Five things I built

01 Nimantrana Digital invitations for Nepali events
02 Personal automations Frost scout
03 Kinmel App Tell it what's in your fridge; gives you 3 recipes
04 Real estate chrome extension Listing-to-client matching for buyer's agents
05 A second brain OpenClaw + Obsidian-style vault + Slack

01

Nimantrana

...means invitation in Nepali

Digital invitations for Nepali events

The "he works in IT" Problem

  • The Request: "Can you design our wedding invitations?"
  • The Reality: I'm not much of a designer. But I have a buddy named Claude...
  • The Assist: Pasted several examples into Claude and asked for templates.
  • The Result: Built a full SaaS. https://nimantrana.online

Traditional
Invitations

Nepali invitations have a very specific look and color.

Red, gold, and honorifics.

Solving for a Niche Audience

  • Two Calendars: Nepali lunar calendar (2082 BS) alongside the standard (2026 AD) calendar.
  • No Special Keyboards: Type in English characters; the app converts it to Nepali script in real-time.
  • Beyond Weddings: Templates for traditional ceremonies like the "Rice Feeding" (Pasni).
  • Proper Etiquette: Automated "Mr. & Mrs." style honorifics for elders.
  • Where the users are: One-tap sharing to WhatsApp.

Tech

nimantrana.online

Tech & Cost

FrontendAstro + SvelteFree
BackendFastAPIVPS
DatabasePostgreSQLSame VPS
HTTPSCloudflare TunnelNo nginx

Total: price of two coffees

Demo Video

nimantrana.online

Product

The Final Result

Real Invitation Demo

Guest Experience

  • Google Maps venue link
  • Real-time countdown timer
  • Instant RSVP feature

Host Control

  • Admin dashboard
  • Guest list management
  • Track attending vs. declined
Easy CRUD app. Took about one weekend.
Helped family use it for a wedding invite.

02

Personal automations

Tiny VPS + Prefect flows + Slack

Problem

  • "Is there frost on my car?" — Check the weather app every morning. Try to figure out whether car is frozen.

Demo

Slack in action

  1. Frost alert from this morning
Luki Bot in Slack

Screenshot — running on personal Slack workspace

Tech & Cost

OrchestrationPrefectWith dashboard to manage flows
AIOpenAI gpt-4o-miniFor all things that need an LLM
DataPostgreSQLFor whatever needs storage
NotificationsSlack WebhooksEverything posts to Slack

Shared hosting with nimantrana.online. OpenAI: pennies per day.

Total time spent: maybe four hours.
They've been running for months.

03

Kinmel App

Tell it what's in your fridge; gives you 3 recipes

The Friction

  • Staring into the fridge at 6 PM.
  • Having random ingredients but no clear recipe.

Demo

Kinmel App in Action

I fully recognize this is just a ChatGPT wrapper...
but it's convenient.

Plus, it gave me an excuse to mess around with iOS development, which was fun.

The Pipeline

Voice Dictate

"I have eggs, spinach, and feta."

OpenAI API

Hits GPT-4o with context

JSON Recipe

Structured data returned

Recipe Cards

Pretty UI displays options

Pantry Management

What's built already:

  • Inventory Backend: PostgreSQL schema tracking every item and its 'bought on', 'best by' dates.
  • Automation Ready: I already use Prefect + Slack for other projects; the 'pipes' are laid.

The Future: Proactive Kinmel

  • "Spinach going bad!": AI notices spinach expires in 2 days and suggests recipes to use it up.
  • Geofenced Reminders: Slack ping as I'm leaving the office: "Hey, you're near Target and low on eggs. Stop and grab a dozen?"
  • Easy data entry: Scan receipt from Kroger

04

A Chrome extension
for a buyer's agent

Started as a tool for myself, then built it out as an MVP

Inside the mind of a buyer's agent

"I find a listing. Then I mentally run through my clients..."

Would the Garcias like this?

Is this in Mike's budget?

Enough bedrooms for the Wilsons?

Wait! Does it have an HOA?

10 new listings/day × 10 clients = 100 mental comparisons
Basically O(n2)

How It Works

  1. Right-click listing (Zillow, Redfin, RealTracs) — "Who wants this house?"
  2. Extension scrapes the page — price, beds, baths, features
  3. Matches against client preferences — budget, location, dealbreakers
  4. Ranked results + one-tap share — match score per client, send a text (FUTURE!) right from the browser

The pipeline I'm building towards

external Listing Page right-click 1 Scrape raw HTML 2 Parse partial JSON 3 LLM Fill listing JSON, parse images chrome.storage.local Client Prefs prefs JSON 4 Match scores 5 Rank sorted 6 Draft message output SMS

Demo

Right-click → Results

Pre-recorded (set speed to 0.5x in player)

The Entire Codebase

background.js

Context menu + scoring logic

~170 lines

content.js

Page scraper + results overlay

~220 lines

styles.css

The overlay UI

~260 lines

Total: ~600 lines. No backend. No database. No server.

05

A personal assistant
that remembers things

OpenClaw + Slack + a folder of markdown files

The problem

I go through moments each day where I want to take notes on something, but I end up not doing it.

  • Voice Dictation: I want to just talk and have it saved.
  • Easy Search: "What treats did I buy the dog last time from Costco?"
  • Recall: "What was that taco place we liked in Germantown?"
  • Knowledge: "Where did I store the spare keys to my car?"
Willow the dog

How it works

InterfaceSlack: I text it in a stream of consciousness manner
AgentOpenClaw in Docker + gpt-4o-mini (for now)
MemoryA vault of markdown files: inbox/, areas/, resources/, archive/
Recall"Remember this" saves a note. Questions check the vault first.
(FUTURE!) TasksWrite scheduled reminders and one-shot tasks via prefect

Demo

Remembering
& Recalling

"Remember this..."

Saving a note about Willow

"What did the vet say?"

Recalling a note about Willow

The vault

vault/inbox/

Unsorted notes

vault/areas/

Ongoing: health, finance, home, dog, travel

vault/resources/

Research, reviews, reference

vault/archive/

Completed or inactive

Markdown files. Auto-committed to git.

What's next

Now File-based search — works well with a clean folder structure
Next RAG over the vault — embed notes, vector search for better recall
Really Next Graph Database — (with some cool UI)
It's just search over markdown files.
But it remembers things I don't.

What I Learned

  1. Pick a small annoyance.
    If it bugs you daily, it's a good candidate.
  2. Boring tech is better.
    I use the same stack as my day job so I can move faster.
  3. AI is just a tool.
    The problem doesn't care if you're using GPT-4 or a regex, as long as it's solved.
  4. Don't let AI drive 100%.
    I want to steer the $ decisions, and high level architecture.
  5. Build it so you can fix it later.
    Can you debug this in 6 months when you've forgotten how it works?

The Cost of All of This

Hetzner VPS$7.59 per month
Cloudflare Pages + TunnelFree
OpenAI API callsPennies/day
Chrome Extension$0
Total$7.59 per month

Low risk to experiment.

Why bother building "tiny" software?


Scratch the itch: Go from idea to MVP in an afternoon for that sweet, sweet hit of dopamine.

Testing the hype: Where does AI *actually* fail?

Stay sharp: I can learn a new framework or pattern in a weekend.

That's it.

Five problems, five side projects, minimal cost.



Ashish Acharya · Knowledge Jam 2026