How a Pune-Based EdTech Startup Cut AI Costs by 60% and Built a Scalable Content Engine
A fast-growing EdTech platform was burning ₹12L/month on AI API costs with a 22% error rate on generated content. We optimized their model strategy and built custom question generation skills that cut costs by 60% while improving accuracy.
Client
Series A EdTech platform, 45K students, JEE/NEET/CAT prep
Industry
EdTech
Timeline
10 weeks including model optimization and testing
The Challenge
The client's AI-powered content engine was their core product, but costs were spiralling and quality was inconsistent. API spend was growing 12-15% month-over-month, and nearly a quarter of generated questions had errors.
- ₹12L/month in AI API costs, growing 12-15% MoM
- 22% error rate on AI-generated questions requiring manual review
- Single-model approach using expensive GPT-4 for all tasks
- Content team spending 40% of time correcting AI output
- Doubt resolution system slow and often inaccurate
The Solution
We implemented a tiered API strategy that routes tasks to the right model based on complexity, and built custom fine-tuned skills for question generation and doubt resolution specific to Indian competitive exam formats.
- Tiered model strategy: lightweight models for simple tasks, premium models for complex reasoning
- Custom question generation pipeline calibrated to JEE/NEET/CAT patterns
- Fine-tuned doubt resolution system with syllabus-aware context
- Automated quality scoring to catch errors before human review
- Caching layer for common question patterns and explanations
- Cost monitoring dashboard with per-feature spend tracking
The Results
Monthly AI Costs
₹12L → ₹4.8L
Question Accuracy
78% → 94%
Content Velocity
Baseline → 3x output
Cost Growth Rate
12-15% MoM → Flat
Want results like these?
Every project starts with a conversation. Tell us about your business and we'll tell you what's possible.
Book a Discovery Call