Palimio

Built by experts. Powered by research.

Palimio combines deep expertise in social media strategy, data engineering, and applied machine learning into a single end-to-end intelligence platform. Every component is purpose-built.

Our team brings backgrounds in data science, NLP research, and content strategy for some of the largest media brands in the world. The platform reflects years of domain expertise distilled into automated, scalable pipelines.

System Architecture

Data Sources

Extensive historical data warehouse + live real-time feeds

Processing

Feature Engineering

Content Tagging

Normalisation

ML / Analytics

Embeddings

UMAP + HDBSCAN

Correlation Engine

Intelligence

Shift Detection

Gap Analysis

Strategic Synthesis

Presentation

Dashboard

PDF Reports

API

Our data engineering pipeline

01

Stage 01

Automated Data Ingestion

You enter your social media handles. We do the rest.

Enter your social media handles and the competitors you want to track. Our pipeline takes over from there, ingesting every post, normalising metadata across platforms, and storing it in a structured data warehouse optimised for analytical queries.

Under the hood

  • Daily ingest. New posts are ingested continuously
  • Cross-platform schema normalisation (timestamps, engagement metrics, content types)
  • Scalable cloud data warehouse with partitioned tables for fast analytical reads
  • Multiple gigabytes of data ingested, processed, and made query-ready
02

Stage 02

Content Analysis & Feature Engineering

Turning raw posts into structured creative signals.

Every post passes through a multi-stage feature engineering pipeline. Our proprietary models analyse the content itself (not just the metadata), extracting six distinct creative dimensions that define the 'DNA' of each piece of content. This goes far beyond hashtag counting or keyword matching.

Under the hood

  • Six engineered dimensions: format, topic, emotional tone, opening hook, scene composition, production style
  • Multi-modal analysis pipeline (video, image, caption, audio and metadata signals combined)
  • Hierarchical classification with confidence scoring and edge-case handling
  • Continuous model evaluation against human-labelled validation sets
03

Stage 03

Semantic Embedding & Clustering

Mapping content in high-dimensional space.

Each post is projected into a high-dimensional vector space based on its content profile. We then apply dimensionality reduction (UMAP) and density-based clustering (HDBSCAN) at two granularity levels to reveal the natural theme structure of your content library, and your competitors'.

Under the hood

  • Dense vector embeddings capturing semantic similarity between posts
  • UMAP projection for interpretable 2D visualisation (Content Galaxy)
  • Two-tier clustering: macro themes and micro niches detected automatically
  • Stored cluster centroids enable real-time assignment of new posts without re-clustering
04

Stage 04

Competitor Benchmarking Engine

Fair, normalised, dimension-level comparison.

Competitor data flows through the same analysis pipeline as your own, ensuring a true apples-to-apples comparison. Our benchmarking engine normalises sample sizes, computes per-feature performance deltas, and identifies the specific creative gaps where competitors outperform you, ranked by impact.

Under the hood

  • Sample-size normalisation: primary account volume matched to competitor scrape depth
  • Per-dimension performance breakdown. The actual video, image, and carousel components are compared and matched to what your competitors are producing. We find EXACTLY what's working and where the gaps are.
  • Content Gap algorithm: ranks content by all dimensions and features, anchored against median views and engagement delta across you vs. competitor pool
  • Per-competitor filtering so you can isolate any individual rival's strategy
05

Stage 05

Intelligence & Strategic Analysis

From patterns to prescriptions.

The processed data is fed through a structured intelligence layer that synthesises trends, detects performance shifts, and generates strategic recommendations grounded in your actual numbers. Every insight is citation-backed. No generic advice, no hallucinated statistics.

Under the hood

  • Time-series shift detection with configurable sensitivity thresholds
  • Engagement correlation analysis across all metric combinations
  • Structured output: 'What's Happening' narrative + 'Recommendations' with named targets
  • Executive summary generation that covers performance, production mix, cadence, and competitive positioning
06

Stage 06

Interactive Dashboard & Reporting

Research-grade analysis, consumer-grade interface.

Everything arrives in a real-time dashboard designed for strategists, not statisticians. Every chart is interactive, every breakdown is filterable, and the full report exports to a pixel-perfect A4 PDF in one click. The interface is built to surface insight fast. No training required.

Under the hood

  • Server-rendered Next.js application with sub-second filter switching
  • Per-platform intelligence with engagement formulas tuned to each platform's mechanics
  • One-click PDF report with AI executive summary, KPIs, top posts, and competitor snapshot

Want to go deeper?

We're happy to walk through the methodology, model architecture, and data handling practices in detail.

Talk to us