METU CENG Senior Project · Group 16 · 2026

The METU-only marketplace that lives where you do.

Items, housing, services, and the chat to close every deal — all behind a verified @metu.edu.tr sign-in. One Kotlin codebase. Android and iOS. No more 400-message Telegram threads.

Watch the demo Explore features

2: platforms · 1 codebase
3: listing kinds
ML: hybrid search + recommender

Home feed showing recommended listings and category chips

Chat thread with a structured offer bubble

The problem

A senior's kettle shouldn't take 400 messages.

Today the trade is everywhere and nowhere: a Telegram group for textbooks, a Google Form for roommates, scattered DMs for tutors. Threads lose context. Offers vanish. Newcomers can't find anything.

Hocamdan puts the whole student economy behind one verified-METU sign-in: items, housing, and services. Per-listing chat threads keep negotiations tidy. A hybrid search + ML recommender surfaces the right post for the right person.

Verified @metu.edu.tr identity end-to-end
One thread per listing — no DM crosstalk
Recommendations cold-start with popular-N, then personalise

Hocamdan

Bosch kettle, used 1 semester ₺220 · 3 offers · 1 thread Accepted ✓

Features

Six pillars. One app.

Mobile-first. Compose Multiplatform shares 95% of UI code across Android and iOS. The backend does the heavy lifting on a single CENG VM.

METU-only auth & moderation

Sign-in is gated on an @metu.edu.tr address. Reports are reviewed and moderated to keep the community trustworthy.

Marketplace · Housing · Services

Three listing types with their own posting flows, status pills, and category filters. Items, sublets, tutoring — one feed.

Hybrid search + ML recommendations

Sentence-transformer embeddings for semantic similarity. LightFM with user-feature and item-feature matrices, a deterministic re-ranker, and popular-N cold-start for new users.

Real-time chat

WebSocket-powered. One thread per (buyer, seller, item). Live ✓✓ read receipts, typing indicators, online-presence dots, mutual rating in the header.

Structured offers

Make / counter / accept / decline as first-class message types. Price-history sparkline on every negotiation. Accepted offers auto-generate a sale-confirmation card.

One codebase, two platforms

Compose Multiplatform shared UI ships natively to Android and iOS. ~95% of UI code is shared; only the thin platform shims diverge.

Architecture

Three layers. Honest infrastructure.

A shared Kotlin client speaks REST + WebSocket to a FastAPI service. That service fans out to MongoDB, ElasticSearch, and a LightFM model file living next to it on the CENG VM.

Layer 1 · Client

Mobile

Compose Multiplatform shared UI

AndroidiOS

Layer 2 · API

FastAPI

Auth (JWT) · email verification
Chat WebSocket
Listings & search
Offers & recommender
Moderation & reports

UbuntuNginx

Layer 3 · Data

Storage & ML

MongoDB — users, listings, chat, offers
ElasticSearch — hybrid keyword + vector
LightFM — recommender model file

Tech stack

Tools we actually use.

Kotlin
Compose MP
Ktor
Koin
FastAPI
Pydantic
MongoDB
ElasticSearch
LightFM
sentence-transformers
Nginx
GitLab

Screen tour

See it move.

One feature-length loop and three short ones. Scroll the rail below for the rest of the surfaces.

Featured Feed + recommendations Personalised rail surfaces as you scroll. LightFM + sentence-transformer similarity, popular-N cold-start for new users.