AI / Entertainment / Content Research · Internal product
Content Sense — AI film discovery engine
An AI content intelligence platform that answers natural-language film questions with structured insights — summaries, performance indicators, character analysis, and recommendations via RAG.
- Timeline
- In active development / internal testing
- Status
- In development
- Category
- software
- Industry
- AI / Entertainment / Content Research
<2 min
Research time
Down from 30–40 minutes per query
Testing
Status
Internal active development
Context
Film research was scattered across noisy search results without narrative context or validated insights.
Content Sense combines LLM reasoning with authoritative metadata so creators, marketers, and researchers get reliable answers instantly.
Channels & stack
- Custom Software Development
- AI Agents & Automation
- Next.js
- SvelteKit
- Tailwind CSS
- Node.js
- Python
- PostgreSQL
- Vector database
- OpenAI
- Google Gemini
- AWS
- Railway
- GCP
Challenge
Hours wasted searching scattered film sources
Noisy traditional search without narrative context
No unified AI + metadata platform for film intelligence
Goals
- 01
Natural-language film search with structured responses
- 02
Reduce hallucinations via retrieval-augmented generation
- 03
Ship conversational UI for creators and researchers
Delivery sequence
How the engagement ran.
Phase 01
Foundation
AI-first architecture
Hybrid LLM + metadata retrieval design.
Work
- Prompt structures
- Metadata enrichment
- Vector retrieval
Deliverables
- RAG pipeline
- Conversational UI
Architecture
- 01
Natural language film search
- 02
Metadata enrichment and performance metrics
- 03
Character and story insights
- 04
Related film recommendations
- 05
Structured JSON response engine
Outcome
Research time cut from 30–40 minutes to under 2 minutes
Creators generate insights and comparisons faster
RAG improved factual correctness vs plain LLM answers
Lessons
- 01
Fuzzy matching and structured prompts reduced hallucinations on vague titles
- 02
A unified metadata schema prevented multi-source inconsistencies
- 03
Continuous prompt tuning from user feedback was essential