InsightForge — Agentic Assistant for Content Creators
Published:
Overview
InsightForge is an agentic assistant built for content creators. It automates the full content production pipeline — from discovering trending topics to generating scripts and publishing across platforms — reducing creator workflow time by 40%.
Built during GDGoC Hackathon 2026 where the team placed Top 2 / 300 teams (Finalist).
Architecture
The system uses a Stateful LangGraph graph as the orchestration backbone, with specialized sub-agents communicating via the A2A (Agent2Agent) Protocol:
- Trend Discovery Agent: Crawls and aggregates trending topics across platforms using MCP-connected tools.
- Script Generation Agent: Produces structured content outlines and full scripts grounded in trending data.
- Publishing Agent: Handles cross-platform formatting and scheduling.
Key Design Decisions
- MCP (Model Context Protocol) Servers: Standardized tool interfaces for web crawling, search, and content APIs.
- A2A Protocol: Ensures type-safe JSON payloads between agents, enabling seamless state and data exchange.
- Structured Output Validation: LLM backbone outputs are validated against Pydantic schemas before being consumed downstream.
- Multi-LLM Support: Pluggable backbone supporting Gemini, GPT-4o, and Claude.
Tech Stack
| Layer | Stack |
|---|---|
| Orchestration | LangGraph (Stateful) |
| Agent Communication | A2A Protocol |
| Tool Access | MCP Servers |
| LLM Backends | Gemini / GPT-4o / Claude |
| Language | Python |
System Architecture
The system orchestrates three specialized agents:
- Trend Analysis Agent: Real-time trend discovery across Google Trends, TikTok, and Threads
- Content Architect Agent: Script generation with structured outlines and multimedia cues
- Posting Agent: Cross-platform formatting and automated scheduling
Workflow
User queries are routed through a Stateful LangGraph orchestrator that:
- Analyzes trending topics
- Generates structured content
- Requests human approval before publishing
- Executes publishing with feedback loops for continuous refinement
Recognition
Impact
- 🏆 GDGoC Hackathon Vietnam 2026 — Finalist, Top 2 / 300 (AI Content Architecture track)
- Reduced content creator research-to-script workflow time by ~40%