Commit 769383

2025-10-08 12:13:03 Mayur Mahajan: Updated the challenges and tech stacik
Projects/Social Media Manager.md ..
@@ 31,46 31,41 @@
* **How does it work?**
1) High-level architecture (components)
* Client/UIs : Web dashboard (React/Vue)— for content creation, calendar, analytics.
- * API Gateway & Auth : OAuth2 / OpenID Connect for social accounts and users, rate-limiting, API key management.
- * Orchestration / Microservices : Microservices for: Content Composer, Scheduler, Publisher Connector, Analytics, Moderation, Media Processor.
+ * API Gateway & Auth : OAuth2 / OpenID Connect for social accounts and users, API key management.
+ * Microservices : Microservices for: Content Composer, Scheduler, Publisher Connector, Moderation, Media Processor.
* Message Bus / Event Stream : Kafka / RabbitMQ for decoupled workflows (scheduling events, publishing events, analytics pipeline).
* Datastores : Relational DB (Postgres) for metadata and users.
- Blob/object store (S3) for images, video.
- Time-series DB / OLAP (ClickHouse, BigQuery) for analytics.
- Redis for caches, locks, rate limiting.
- * Model Serving : managed endpoints (OpenAI, Hugging Face Inference).
- * Third-party Connectors : Social platform APIs (Facebook/Meta, Instagram, X/Twitter, LinkedIn).
+ Blob/object store (S3) for images.
+ Redis for caches.
+ * Model Serving : managed endpoints (OpenAI, Gemini).
+ * Third-party Connectors : Social platform APIs (Facebook/Meta, Instagram, LinkedIn).
* Human-in-the-loop : Review queues for moderation, creative approvals.
2) Data flow (step-by-step)
- * User creates content in the dashboard or requests AI-generated content (text, image, video).
+ * User creates content in the dashboard or requests AI-generated content (text, image).
* Pre-flight validation & moderation : Content passes through safety filters (automated moderation models + policy rules). If flagged → human review queue.
- * Content generation & enrichment (if requested): Text generation (caption, hashtags), image generation/edits, video clips, or audio transcribe.
+ * Content generation & enrichment (if requested): Text generation (caption, hashtags), image generation/edits.
Enrichment: sentiment tags, target-audience suggestions, best posting times (ML predictions).
* Media processing : Resize, crop, compress, create thumbnails, burn-in captions.
* Scheduling : Scheduler service enqueues job to message broker with publish timestamp and target social connectors.
* Publishing: Worker picks up job, calls social API with proper rate-limit handling and retries, logs response.
- * Monitoring & Analytics : Publisher returns post-id; analytics pipeline ingests impressions/likes/comments via webhooks or polling; stored in OLAP for dashboards.
- * Feedback loop : Engagement metrics feed the recommendation models (which suggest better times, copy variants, or creative). Data used for periodic model retraining.
- * Audit & Compliance: All content, approvals, and publish events stored for audit/GDPR/record-keeping.
3) AI models & technologies (by capability) Content generation (text)
- * Models: GPT-4 / GPT-4o (OpenAI) or Hugging Face Transformer models (Llama2, Mistral, T5 derivatives) for on-prem / fine-tuned workloads.
- * Libraries & infra: OpenAI API for managed LLMs; Hugging Face Transformers (PyTorch/TensorFlow) for self-hosted.
+ * Models: GPT-4 / GPT-4o (OpenAI) or Gemini(vertexAI) for image/text generation and sentiment analysis.
+ * Libraries & infra: OpenAI API for managed LLMs; VertexAI APIs for gemini.
* Use-cases: captions, thread generation, A/B caption variants, content re-purposing, tone/style transfer.
- * Tech notes: Use RAG (retrieval-augmented generation) when injecting brand guidelines or user history into prompts.
* **What makes it innovative?**
- Our Social Media Manager goes beyond simple scheduling by combining generative AI (GPT-4, Hugging Face models) with computer vision (OpenCV, CLIP, YOLO) and predictive analytics. It not only creates brand-safe captions, hashtags, and visuals but also learns from engagement data to recommend the best posting times and formats. This closed-loop system of AI-driven creativity, moderation, scheduling, and performance forecasting makes it an end-to-end intelligent solution — not just a tool, but a creative and strategic partner.
+ Our Social Media Manager goes beyond simple scheduling by combining generative AI (GPT-4, Gemini) with computer vision (OpenCV, CLIP, YOLO) and predictive analytics. It not only creates brand-safe captions, hashtags, and visuals but also learns from engagement data to recommend the best posting times and formats. This closed-loop system of AI-driven creativity, moderation, scheduling, and performance forecasting makes it an end-to-end intelligent solution — not just a tool, but a creative and strategic partner.
## Technology Stack
* **AI/ML Models:** OpenAI API, gemini-2.5-flash, etc.
- * **Frameworks & Libraries:** LangChain4J
+ * **Frameworks & Libraries:** LangChain4J/LangGraph4j
* **Backend:** Java, Node.js
* **Frontend:** React, Vue.js
* **Databases:** PostgreSQL
* **Deployment & Tools:** Docker, Git, AWS
- * **APIs Used:** Twitter, FB, Instagram, LinkedIn, etc.
+ * **APIs Used:** FB, Instagram, LinkedIn, etc.
## Project Architecture
@@ 87,7 82,10 @@
## Challenges & Learnings
- 1. Challenges & Learnings: Identifying the right plugin for Figma design and learning the basics of Figma screen design.
+ 1. Identifying the right plugin for Figma design and learning the basics of Figma screen design.
+ 2. Generate image using different models as per event.
+ 3. Analyse the sentiments of generated content.
+ 4. Analyse the user/brand backgroung to generate the content.
## Future Roadmap
@@ 110,7 108,7 @@
## Repository & Live Demo
- * **GitHub Repository:** [Link to your code repo here]
+ * **GitHub Repository:** [[Link to your code repo here]](https://gitlab.enovate-it.com/enovate/hackathon/social-media-manager)
* **Live Demo:** [Link to your live demo or video walkthrough here] (Highly encouraged!)
## Screenshots / Demo Video
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9