VaboBD-algorithm

Open-source feed/ranking algorithm for VaboBD — an Facebook-like social platform for Bangladesh and South Asia.

Status: scaffold only. No code yet — the MVP pipeline below is the design to implement.

The pipeline shape is inspired by X's open-source x-algorithm (Home Mixer pipeline concept), but the MVP is heuristic only — no AI/ML. The architecture deliberately keeps the ML scorer as a swappable step 3, so AI/ML can be added later without restructuring the pipeline (REQUIREMENT.md §10).

MVP pipeline

1. Candidate sources (parallel)

   • In-network: recent posts from followed accounts (Postgres)
   • Out-of-network: trending/recent posts from non-followed accounts (Postgres)

2. Pre-scoring filters — already-seen (Redis set per user), blocked/muted accounts, age limit, reported/flagged content

3. Scoring (heuristic)

   score = (w1 × recency_decay) + (w2 × like_count) + (w3 × comment_count)
         + (w4 × is_following) − (w5 × already_seen_penalty)
   

   Weights are config values, tunable like production defaults in x-algorithm.

4. Selection — sort by score, take top N

5. Post-selection filters — dedup, final block/report check

Timing & storage

• Computed real-time, at request time — no background cron/pre-compute.
• Backed by Redis caching (follow list, recent-posts pools, seen-posts set); Postgres remains the source of truth but is not hit on every request.
• Scoring weights and visibility filtering (block/mute/report) mirror Instagram's approach.

Docs

• docs/ — public docs for this repo (design notes, weight tuning guidance) — to be filled during implementation.

Related repos

Repo	Purpose
VaboBD	Root repo — project requirements and docs
VaboBD-web	Next.js frontend
VaboBD-backend	Go backend

License

TBD — see root repo license note.