our work

things we've built
for people who were tired.

Every entry is a real problem, a real build, a real outcome. Replace these with your actual case studies as you ship them — the template stays the same.

4+
Projects shipped
200+
Hours saved/week
< 48h
Response time
100%
Code delivered

Recent projects

showing 4 of 4
CASE / 01
D2C · Mumbai
Instagram DMs to CRM
in under 60 seconds
Two founders drowning in Instagram leads. Now it's handled before their morning coffee.
The problem
Two people were spending 3 hours a day copy-pasting Instagram DM leads into HubSpot, then sending manual WhatsApp follow-ups. Leads fell through constantly.
What we built
Meta Graph API → LLM extraction of lead data → auto-push to HubSpot with tagging → WhatsApp follow-up sequences keyed to lead type.
Python Meta Graph API HubSpot Claude WhatsApp Cloud API
saved ~3 hours/day of manual work
CASE / 02
Recruitment · Bengaluru
Cross-site job aggregator that runs itself
Fired the manual process, not the intern. Reassigned them to actual recruiting.
The problem
A recruiter was paying an intern ₹25k/month to check 20 job boards every morning for fresh listings, then post them to Slack. Errors and missed listings were routine.
What we built
Scheduled Python scrapers for 20 target sites, de-dup layer, filtering rules for their niche, clean push to Slack at 9am daily with full metadata.
Python Playwright PostgreSQL Slack API Cron
replaced a ₹25k/month intern role
CASE / 03
Finance · Pune
Vendor invoices into Tally, automatically
Month-end used to ruin weekends. Now it ruins one afternoon.
The problem
Month-end close took 4 full days. One person had to open 200+ PDF invoices, type vendor data into Tally, and match against POs. Errors cascaded into GST filings.
What we built
PDF ingestion pipeline → Claude-powered extraction of vendor, GSTIN, line items, amounts → PO-matching logic → Tally XML import. Human reviews only flagged exceptions.
Python Claude API Tally TDL pdfplumber
month-end close: 4 days → half a day
CASE / 04
Edtech · Bengaluru
A support bot that actually reads the docs
Their support team stopped answering the same 30 questions. Started doing actual support.
The problem
Support team was answering the same 30 questions over and over. Students wouldn't read the FAQ. Response times were creeping past 6 hours.
What we built
RAG-based support agent trained on their docs and past tickets. Handles ~70% of queries instantly. Escalates the rest with full context pre-loaded for the human.
TypeScript Claude API Pinecone Next.js Intercom
support resolution: 6 hrs → 4 minutes
"We don't sell AI transformation.
We build specific, ugly, effective systems
that replace specific, ugly, expensive problems."
— our whole worldview, on one card

your problem next?

One form. Takes 2 minutes. We reply within 24 hours.

bring us a problem →