PALAWAN CREATOR

AI-Powered Social Media Automation for Filipino Community Creators

Puerto Princesa, Palawan • Taglish • Instagram Reels • Zero Manual Work

What It Is

Palawan Creator is a fully automated Instagram content engine built specifically for Filipino community creators. It handles everything from finding the news to writing the script, generating the voice, assembling the video, and posting the Reel — on a schedule, without human intervention. It runs as a desktop app that any non-technical creator can install, configure once, and let run.

This is not a scheduling tool. It is not a caption generator. It is a complete end-to-end production system that replaces hours of daily content work with a single running process.

The Problem It Solves

Community content creators in the Philippines — particularly in regional areas like Palawan — face a brutal reality: audiences expect daily content across multiple formats and topics, but the production capacity of a solo creator is capped at a few posts a week. The gap between what the algorithm demands and what a human can produce alone is enormous.

Add to that the linguistic complexity of Filipino audiences — who expect content in Taglish, not corporate English — and the regulatory sensitivity of topics like government schemes, job listings, and local news, and the challenge becomes clear. Generic AI content tools are built for Western, English-speaking markets. They produce content that sounds wrong to a Filipino ear.

Palawan Creator was built to eliminate that gap entirely.

Seven Content Pillars. Seven Distinct Voices.

The system operates across seven specialized content pillars, each with its own persona, posting window, and audience:

• BALITA NGAYON — Local news delivered by Manong Ronnie. Morning posts between 6 AM and 9 AM PHT. Straight-talking, community-first reporting voice.

• JOB ALERT — Employment and training opportunities by Ate Cita. Morning window. Validates every listing against 7 trusted Philippine sources before posting.

• OFW PAMILYA — Stories and support for overseas Filipino workers and their families, by Kuya Jun. Evening posts optimized for OFW viewing hours.

• LUTONG PALAWAN — Local cuisine and recipes by Nanay Mely. Lunch-window posts designed for peak food-content engagement.

• HIWAGA KALIKASAN — Nature and science by Guro Dante. Afternoon educational content with a curious, teacher's tone.

• KALIKASAN PALAWAN — Environmental conservation by Ate Issa. Afternoon to evening, focused on Palawan's ecological identity.

• GOV SCHEMES — Government programs including 4Ps, OWWA, and TESDA by Manang Nena. Wide posting window covering the full working day.

Each persona has its own vocabulary, sentence rhythm, code-switching style, and emotional register. The audience does not experience a brand — they experience a neighbor.

The Automation Pipeline — End to End

News Scoring Engine

Six RSS feeds are aggregated and each article is scored across four dimensions: Local Relevance, Pillar Fit, Engagement Potential, and Content Safety. Only articles that pass a composite threshold proceed to content generation. No human selects the topic — the engine does.

Script Writing with Constitutional Safety

Once a topic is selected, the LLM writes a script in the voice of the assigned pillar persona. Before that script is ever used, it passes through a five-gate constitutional AI safety stack:

• Gate 1 — Factual Accuracy: claims are verified against source material

• Gate 2 — Name Removal: private individuals are anonymized

• Gate 3 — Political Filter: partisan content is neutralized

• Gate 4 — Cultural and Safety Review: content is checked against Filipino cultural norms and community safety standards

• Gate 5 — TTS Readability: text is optimized for natural voice synthesis

A script that fails any gate is either revised or discarded. Nothing goes to air that has not cleared all five checks.

Caption Generation

Captions are generated with exactly 25 hashtags, within a 220-character limit, in Taglish. The system understands Filipino hashtag culture — mix of Filipino, English, and local Palawan-specific tags — not a generic hashtag generator.

TTS with Word-Level Timestamps

Text-to-speech uses Microsoft Edge's fil-PH-BlessicaNeural voice — a natural Filipino female voice that handles Taglish code-switching correctly. The system captures word-boundary timestamps during generation, which are used to synchronize burned-in subtitles frame-accurately during video assembly.

Video Assembly via FFmpeg

Stock footage is sourced from the Pexels API and matched to the content topic. FFmpeg assembles portrait 1080x1920 MP4 files — the native Instagram Reel format — with the voice track, background music layer, and burned-in word-synchronized subtitles. FFmpeg is bundled inside the installer; users do not need to install anything separately.

Instagram Posting via Graph API

The assembled Reel is posted to Instagram via the official Graph API v19.0. The system supports both Reel and image post modes. Media is hosted via catbox.moe for the Graph API upload flow, removing the need for the creator to manage hosting infrastructure.

Hidden Technical Strengths

Multi-Provider LLM Chain with Fallback

The LLM dispatcher routes across four providers in configurable order: Gemini 2.0 Flash (1M-token context, 60 RPM free), Groq (Llama 3 / Mixtral, ultra-fast inference), OpenRouter (aggregated model access), and Pollinations.ai as a zero-key fallback. If any provider fails or rate-limits, the next takes over transparently. The pipeline never stalls because one API is unavailable.

Provider order is cached in-memory for 60 seconds to avoid database overhead on every LLM call — a small optimization that matters at high cadence automation.

SQLite WAL Mode with Shared Connection Proxy

The database uses SQLite in WAL (Write-Ahead Logging) mode, allowing concurrent reads during writes — critical for a system where the scheduler, API routers, and voice assistant all access the database simultaneously. A single shared connection proxy is used across 47+ call sites instead of opening new connections per request, eliminating overhead and connection contention.

Topic Pool Pre-Generation

Rather than scoring news in real-time at posting time, the system pre-generates a topic pool every 72 hours. This means posting never waits for RSS fetching and scoring to complete — topics are already ranked and queued. Posting latency is near-zero.

APScheduler with Missed-Post Recovery

The scheduler runs in PHT (Asia/Manila) timezone with pillar-specific posting windows. If a post fails — network timeout, API error, FFmpeg crash — the system logs it and automatically retries within a 4-hour recovery window. No post is permanently lost to a transient failure.

Breaking News Detection

The system monitors RSS feeds continuously for trending topic spikes. When a breaking news signal is detected, it auto-triggers the relevant pillar's pipeline outside the normal schedule. The creator does not need to be at their desk for the account to respond to breaking local news.

Voice-Controlled Pipeline

A WebSocket-connected voice assistant lets the creator control the pipeline by speaking. Filipino intent classification maps natural language commands — in Taglish — to specific pipeline actions. Say 'i-post na yung balita' and the system starts a BALITA_NGAYON run. This is designed for a creator who may be out in the field, not at a desktop.

Fernet Encryption for API Keys

All API credentials are stored encrypted using Fernet symmetric encryption. The encryption key is stored outside the application directory — in %APPDATA%/PalawanCreator/data/ — so even direct file access to the application folder does not expose credentials.

Password-Gated Niche Changes

Changing the account's content niche requires a 4-digit PIN. This prevents accidental misconfiguration — a single wrong tap in the settings UI cannot silently reorient an account's entire content strategy. It is a small but deliberately considered design decision for non-technical users.

Multi-Account Architecture

The system supports multiple Instagram accounts, each with its own niche, pillars, LLM provider preference, TTS voice, posting schedule, and posting history. A creator managing multiple community pages — or an agency managing clients — runs them all from one installation.

Zero Operational Cost Model

The entire system runs on free-tier APIs at sustainable volume:

• Gemini 2.0 Flash — 60 requests per minute, 1 million tokens of context, free

• Groq — free tier with Llama 3 and Mixtral, fastest available inference

• Edge TTS (BlessicaNeural) — no API key required, hosted by Microsoft

• Pexels — free stock footage API with generous rate limits

• Pollinations.ai — zero-key image generation, no account required

• Catbox.moe — anonymous media hosting for Instagram Graph API uploads

• Instagram Graph API — free for creators using their own account token

A creator running Palawan Creator at full capacity — multiple posts per day across seven pillars — pays nothing in API costs. The only cost is the machine it runs on.

Deployment

• Single Windows .exe installer — PyInstaller bundles Python backend, FFmpeg, and all dependencies

• Frontend served by Electron — no browser required, no localhost URL to navigate to

• First-run setup wizard collects API keys and Instagram token — everything else is automated

• User data stored in %APPDATA%/PalawanCreator/data/ — isolated from the app, survives reinstall

• Auto-updater built in — new versions deploy silently without creator involvement

Palawan Creator is not a content tool.

It is a content operation — built for one creator, scaled for a community.