A selection of automations, agents, and apps we built to sharpen the craft. Some were prototyped for sectors we wanted to learn cold. One is a personal project we still use. All of them taught us how to ship faster for the next one.
Featured Build · 01
An AI study assistant built for exam season.
A personal project. Claude-inspired UI, a vector database of 65 PDFs, four modes of working. The most polished of the set, and the one still in active use.
Personal Project
Subjects
Question
Explain the difference between above-the-line and below-the-line marketing with examples.
1m4m8m · selected
Structured answer · 8 marks
ATL targets a mass, undifferentiated audience; BTL targets specific segments with direct, measurable activity.
TTL combines both for integrated campaign coverage.
Tell me about
Why do brands invest in content marketing over direct advertising?
Walkthrough
Trust is the bottleneck, not reach. Content marketing builds credibility over time — a useful blog post or video earns repeat visits, while a paid ad disappears the moment the budget does. The CAC drops as organic authority compounds.
Go deeper
Quick quiz · Digital Marketing · 1 / 5
Which metric measures the percentage of users who click a link after seeing it?
Batch · 12 questions processed
01Define the marketing mix and its four components.
02Compare inbound vs outbound marketing strategies.
03What is brand equity and how is it built over time?
65PDFs in vector DB
4working modes
5subjects, separate memory
.docxbatch export, one click
Built this the month before my own university finals, partly to study and partly to see how far a single person could push a Claude-style app in a weekend. Kept it open for a few friends. It still gets opened every exam season.
More from the bench
Smaller builds. Real mechanics.
Each of these is a working prototype: built end-to-end, not a slide deck. Filter by what you care about.
02 / HospitalityWhatsApp Automation
Table reservation bot
Captures bookings from WhatsApp, parses out the time, name, party size, and any special asks like a baby seat. Sends a reminder before the slot and a feedback message after.
Hi, table for 4 at 8pm tomorrow, need a baby seat please.
Confirmed · Sat 8:00pm · party of 4 · baby seat noted.
Reminder · your table is 2 hours away.
How was your evening? 30 seconds of feedback?
Logged to: a Google Sheet with name, date, time, party size, special instructions, feedback, status.
03 / HospitalityWeb App + Workflow
Reputation management system
Three pipelines running at different cadences. Daily review scrape, weekly theme detection, quarterly sentiment trend. The owner gets a WhatsApp digest, the rest lives in a web dashboard.
Daily
Weekly
12-Week
The interesting bit: the 12-week pipeline replays metrics it stored weekly, so trends are real data, not snapshots.
04 / HospitalityWorkflow
Daily sales summary
Pulls a day's orders, cross-references the product catalogue, writes a plain-English summary for the owner on WhatsApp. Built standalone; clearly more useful as a Petpooja integration layer than as a replacement.
Honest take: most cafes use Petpooja, which already does basic summaries. The real value is plugging this on top to turn raw reports into a daily plain-English message.
05 / HospitalityWhatsApp Automation
Customer win-back
Notices when a regular has gone quiet, pulls what they ordered last, drafts a warm WhatsApp message with the right details slotted in. Approved templates with AI-written variables underneath.
Hey {name}, we miss you. Still have that {last_order} with your name on it. Been {days_since} days. Come back soon?
→ "Hey Priya, we miss you. Still have that hazelnut latte with your name on it. Been 21 days. Come back soon?"
Where this goes next: a building block in a wider loyalty system, not a standalone product.
06 / EducationAI Agent
Admission inquiry agent
A chatbot that knows a coaching center's offerings cold and answers parents and students with the same patience at 11pm as at 11am. WhatsApp-ready. Built to plug into the channel the center already uses.
Live modeParent: What batches start in June?Agent: Two JEE batches and one NEET.
Training modeAI as customer: Do you offer demo classes?You answer · AI scores reply
The interesting bit: a two-way training mode lets the owner roleplay with the AI to tune answers before going live.
07 / B2BWorkflow
Lead generation pipeline
Scrapes Google for a target type of business, then runs each result through configurable qualifiers before it ever lands in the sheet. The qualifiers are the product: LinkedIn presence, valid website, email visible.
Apify scrape · raw results1,243
Gate · LinkedIn present? cross-reference412
Gate · valid website? crawl + verify128
Qualified · → Google Sheet47
What makes it useful: the qualifiers are configurable per run, so the same pipeline becomes a different list with one toggle.
What's next
Got a workflow that looks like one of these?
The next one we build could be yours. Tell us what's eating your week and we'll come back with a plan.