Case Studies

Real systems for real businesses

Every project starts with understanding how your business runs. Here's what that looks like in practice.

New Day Learning

Healthcare / Medicaid Services

Case Study 01

Staff management portal for a Medicaid service provider in New Jersey

The Problem

New Day Learning LLC provides DDD services — life coaching, transportation, music therapy — to individuals with developmental disabilities in New Jersey. Every client engagement starts when DDD issues a Service Detail Report PDF with Medicaid info, authorized services, weekly allocations, and rates. Staff were reading these PDFs, copying fields into spreadsheets, and tracking sessions on paper. Billing meant matching completed sessions against authorization limits by hand, then building invoices manually. Client data, authorizations, and session logs all lived in different places.

What I Built

A staff portal that connects the full workflow: PDF import, client management, session tracking, and invoicing. Staff uploads a DDD Service Detail Report — the system parses it, creates the client record, builds the authorization (PA number, date range, unit type, rate), and generates the weekly session schedule with staff assignments. Staff log completed sessions each week. Invoices are generated directly from completed sessions, with payment status tracked per invoice. The portal also manages the service catalog and handles staff authentication with role-based access via Clerk.

Zero

Manual data entry for new authorizations

1 system

Replaces scattered spreadsheets

Automated

Invoice generation from sessions

Stack

Vite + ReactExpress APIPostgreSQL 16Clerk AuthDokployTraefik

GSE Intelligence

Financial Services / Capital Markets

Case Study 02

AI market intelligence platform for institutional investors on the Ghana Stock Exchange

The Problem

Ghana's stock exchange has 37 listed securities and no Bloomberg equivalent. Institutional investors — fund managers, licensed dealing members, research firms — pull data from scattered sources: Excel files from the GSE website, monthly PDFs, news across six business outlets, and their own spreadsheets. Getting a clear market picture takes hours of manual assembly every morning. Producing research is worse — a company analysis means pulling fundamentals from one source, price history from another, macro data from a third, then writing it up by hand.

What I Built

A platform with 10 dashboard pages, 15 API routers, 50+ endpoints, and 14 scheduled tasks. The data pipeline ingests from 7 sources automatically — prices poll every 5 minutes during trading, EOD data upserts at close, fundamentals refresh weekly, monthly GSE PDFs get parsed using Claude, news from 6 sources refreshes every 15 minutes with auto-tagging, corporate actions scrape daily. The AI layer generates daily and weekly market reports, single-stock deep dives with LLM verdicts, and runs a deep research engine with three parallel sub-agents that converge through a critique pass. IC memos generate as 6-section DOCX files. Every night, 12 priority tickers run through parallel research that adds to vector memory. Analysts can ask questions through an agentic chat with 6 tools.

7 sources

Automated data ingestion

Minutes

Research that took a full day

Nightly

Platform gets smarter over time

Stack

Next.js 15FastAPIPostgreSQL + pgvectorRailwayOpenRouterLangGraphCelery + Redis

FarmOps

Agriculture / Poultry

Case Study 03

AI-powered operations platform for a remotely managed poultry farm in Ghana

The Problem

6,000 layers, broilers coming soon, 15-person team on site — and the owner lives in a different city. Feed gets purchased but nobody tracks daily issuance per house, so there's no way to know if usage matches what the birds need. One house got 20% more feed than necessary for six months before anyone noticed — the equivalent of a full batch's profit in wasted feed. Workers report egg counts over WhatsApp photos and voice notes, but nothing connects to what's sold or what cash was collected. The farm manager gives updates on a weekly call, which means the owner finds out about problems days after they start.

What I Built

A platform for 10-20 users with AI wired into every data entry, alert, and report. Workers submit daily data through mobile forms — Claude Haiku flags anomalies before submission ("You reported 420 eggs vs 580 yesterday. Is this correct?"). Every form requires photos, every submission logs who and when, records can't be modified without a justified audit trail. The owner gets a Telegram briefing at 7am — a plain-English narrative of what happened, what's wrong, what needs a decision. Feed variance above 10% triggers same-day alerts. Egg reconciliation tracks collected → stored → sold → cash. 14 SOPs are digitized as mobile checklists with AI completion scoring. Purchase approvals above a threshold route to the owner's phone.

10 min/day

Owner manages farm from phone

Same-day

Feed waste detected and flagged

$10-15/mo

Total AI cost

Stack

Supabasen8nNext.jsReact NativeClaude Haiku + SonnetTelegram Bot

Want results like these?

Every project starts with a conversation. Tell us what's not working and we'll tell you what we'd automate.