Snapshot
Edtech platform for the Feyntech edtech arm. Course delivery, scheduling, and a small enrolment funnel layered on Flask. Since retired; this case study documents the architecture as it was built and shipped.
Context
Feyntech.AI in-house. The Edugator side of the business runs cohort-based courses (Indian high-school and competitive-exam prep). They needed a single surface that handled discovery, enrolment, and the schedule - not three separate tools glued together.
Problem
A cohort-based course platform has a different shape than a Coursera-style library:
- Enrolment is a window, not a click. Cohorts have a start date, a cap, and a deadline.
- Schedule changes ripple - moving a class moves the cohort, not a video.
- The prospective student's parent is also a stakeholder. The marketing surface has to satisfy both.
- The platform has to work on phones, on bad networks, in browsers that aren't always up-to-date.
Approach
One Flask app, three jobs. Marketing pages, an enrolment funnel, and a scheduling surface - all in one deployable. The cost of an internal microservice split would have been higher than the benefit for the team's size.
Cohort as the primary noun. The data model centers on the cohort: it has a start, a cap, a roster, a schedule. Classes, materials, and notifications hang off it. That makes the operator's job (running an actual cohort) the cheap path.
Phone-first marketing surface. Most landing-page traffic comes from WhatsApp shares. The hero loads in under a second on a 3G phone - most of the markup is server-rendered, the JS is the parts that actually need it.
Enrolment funnel. A multi-step form that's resilient to mid-flow drop-off - the data persists at every step, so a parent who reads a question to a kid and comes back five minutes later doesn't lose the form.
Tech
- Python 3 + Flask
- Postgres
- Tailwind on the marketing pages, light Vanilla JS on the interactive bits
- Replit deployment
What I learned
Edtech in India is a parent product as much as a student product. The marketing copy has to read in two registers - the kid skims, the parent decides. That changed the hero structure and the FAQ structure more than any technical decision I made.
