The web and iOS product, September 2026. Screens show the product at that time.
iOS + Android downloads
3,200+
Registered accounts
2,200+
Published event records
15,000+
Product snapshot / Event records include past and upcoming events. Downloads and accounts measure different things.
01 / Bodrum Flow
The product and the work around it
People use Bodrum Flow to explore events, venue programmes and experiences. Building it means working across the mobile app and website as well as the less visible tools that keep event information usable.
As founder and developer, my work spans product interfaces, structured data, AI-assisted processing and operational tooling. The examples below come from the implemented systems behind the product.
One product, several surfaces
The React Native and Expo app supports saved plans, venue following and direct event actions. The Next.js web product provides an immediate way to discover and share programmes. Content is presented in Turkish, English, German and Russian.
Real product screens captured on 5 September 2026. Dates and availability belong to that capture.
02 / Behind the interface
From source material to useful information
Social posts, Stories and submitted media contain information in different forms. The content workflow keeps source context alongside structured event records.
01
Capture
Source posts, Stories and submitted posters or videos.
02
Interpret
OCR and multimodal extraction, with the source context.
03
Validate & review
Date and identity checks, guarded updates and review paths.
04
Use in the product
Structured records for discovery and editorial distribution.
Review and correction remain part of the workflow. Automated processing does not imply that every result is ready to publish.
03 / Engineering
Decisions inside the system
Extraction & data quality
Reading a poster is only the first step
The BF_InstaScrab ingestion pipeline combines OCR, image context and multimodal extraction. Deterministic rules then normalise dates, programme layouts and event identity, retaining the evidence behind the structured result.
For Stories, publication time and capture time are separate. A reference such as ‘tonight’ must be grounded in source identity and timing evidence. Unresolved dates can be flagged for review.
Event identity
Guarding against the wrong merge
Matching venue, day and time is not enough to move media between events. Performer conflicts or uncertain identity can block a transfer, and a duplicate-review label alone does not authorise it.
Allowed merges use guarded database operations with update preconditions and lineage. Locked records stay protected. The aim is to avoid silently attaching another event's poster or video.
Reliability
Separating capture from processing
A durable local spool stores captured work before processing. Atomic writes and job claims preserve ownership, while recovery logic can requeue abandoned work when the available timing evidence is reliable.
Retries have increasing delays and limits. These mechanisms let ingestion resume after interruptions without treating every failure as a reason to start again.
AI & editorial controls
Model output inside a controlled workflow
The moderation system combines image and text evidence with structured model output. Code then checks venue, date and conflict rules. A proposed publication can be held for review when those checks fail.
Review fingerprints and a durable ledger keep unchanged evidence from triggering the same work repeatedly. New evidence or an operator can reopen a case, while record and cost budgets constrain each run.
Intake & distribution
Making room for how people work
FlowBot, the WhatsApp intake service, pairs posters, videos and captions that arrive in different orders, then stores their context in a processing queue. It also preserves captions received while a video is being processed.
A separate editorial interface supports day and night event selections. Editors choose the records; server and database checks reject unpublished, flagged or incorrectly dated entries before saving the selection.
Technology in context
Mobile experienceReact Native / Expo
An iOS and Android app for discovery, saved plans and event actions.
Web experienceNext.js / TypeScript
Public discovery, shareable venue programmes and multilingual pages.
Capture, extraction, validation, messaging intake and operational jobs.
Records & controlPostgreSQL / Supabase
Structured content, guarded updates, queue records and editorial checks.
Continuing the product
The next questions are wider coverage across Bodrum, more relevant discovery and easier updates for local businesses. These are directions to explore, informed by the product and its day-to-day operation.
