Technical Challenges
Personalized video at scale for LATAM financial services isn't just about rendering templates. We're solving hard problems that most platforms ignore or handle poorly.
The Hard Problems We're Solving
1. Spanish TTS with Regional Accents
The Challenge: LATAM Spanish isn't monolithic. Chilean Spanish sounds nothing like Argentine Spanish. Financial institutions need their videos to sound local, not like a generic Spanish translator.
Why It's Hard:
- Most TTS engines offer "Spanish (Latin America)" as a single accent
- Regional prosody, intonation, and vocabulary differ significantly
- Financial terminology varies by country (AFP in Chile vs AFORE in Mexico)
- Emotional tone matters in financial communications (reassuring vs urgent)
Our Approach:
- Fine-tuned TTS models per market (Chile, Peru, Colombia, Argentina)
- Financial services vocabulary optimization
- A/B testing voice characteristics with real clients
- Fallback gracefully when regional models aren't available
Impact: Higher engagement and trust. When an AFP member hears "tu pensión" in Chilean Spanish from a Chilean-sounding voice, it feels personal.
2. Sub-Second Storyboard Generation
The Challenge: Clients need to preview personalized videos before generating thousands of them. Traditional approaches take minutes per preview, making iteration painful.
Why It's Hard:
- Full rendering is slow (transcoding, compositing, effects)
- Data integration adds latency (fetching customer data, calculating values)
- Storyboards need to be accurate enough to trust
- Needs to work with dynamic content (graphs, charts, timelines)
Our Approach:
- Smart caching of static assets
- Lazy rendering (preview frames vs full video)
- Pre-computed templates with variable placeholders
- WebGL-accelerated client-side preview (Vite + lightweight renderer)
- Differential updates (only regenerate changed segments)
Tech Stack:
- Client: Vite + React for instant hot-reloading previews
- API: .NET microservices for orchestration
- Cache: Redis for template/asset caching
- Preview: WebGL compositing for real-time frame generation
Impact: Iterate on video campaigns in minutes, not hours. Clients can test 10 variants before committing.
3. Dynamic Data Integration with Banking Systems
The Challenge: Financial data lives in legacy systems with strict security requirements. Personalized videos need real-time or near-real-time data without compromising security.
Why It's Hard:
- Banks use ancient mainframes (COBOL, AS/400) alongside modern APIs
- PII and financial data require encryption at rest and in transit
- Data freshness vs security tradeoffs
- Compliance logging for every data access
- Handling partial/missing data gracefully
Our Approach:
- Secure data connectors for common banking platforms (Temenos, Oracle FLEXCUBE, SAP)
- Encrypted data pipelines (end-to-end TLS 1.3+)
- Smart fallbacks for missing fields (default values, hide segments)
- Audit trails for compliance (who accessed what, when)
- Data minimization (only fetch what's needed for rendering)
Tech Stack:
- .NET microservices for secure data orchestration
- AWS Secrets Manager for credential management
- AWS KMS for data encryption
- CloudTrail for audit logging
Impact: Videos that feel current and accurate. "Your balance is $12,450" not "Your balance is [loading...]"
4. Multi-Format Rendering at Scale
The Challenge: Same video needs to work on email (MP4), WhatsApp (compressed), SMS (link to web player), and app notifications. Each has size and format constraints.
Why It's Hard:
- Email clients limit attachment sizes (often <5MB)
- WhatsApp compresses videos aggressively
- Mobile data costs matter in LATAM
- Adaptive bitrate for varying connection speeds
- Preview thumbnails need to load instantly
Our Approach:
- Multi-resolution rendering pipeline (4K, 1080p, 720p, 480p)
- Smart compression profiles per channel
- Progressive streaming (start playing before full download)
- Thumbnail generation with meaningful frames (not first frame)
- WebP/AVIF for preview images
Tech Stack:
- AWS MediaConvert for batch transcoding
- CloudFront for CDN delivery
- Lambda@Edge for dynamic format selection
- S3 for multi-format storage
Impact: Videos that work everywhere, look good everywhere, load fast everywhere.
5. Real-Time Compliance Validation
The Challenge: Financial communications have strict regulatory requirements. Videos must comply before sending (no promotional language in pension updates, mandatory disclosures, etc.).
Why It's Hard:
- Regulations vary by country, product type, and customer segment
- Dynamic content makes pre-approval difficult
- False positives slow down campaigns
- Audit trails required for regulators
Our Approach:
- Rule engine for compliance checks (text, voice, visual)
- Pre-flight validation before rendering
- Flagging system (block, warn, audit)
- ML models trained on regulatory documents
- Human-in-the-loop for edge cases
Tech Stack:
- .NET rules engine with configurable compliance policies
- AWS Comprehend for text analysis
- Custom ML models for financial regulation detection
- Audit logs in CloudWatch + long-term S3 storage
Impact: Faster campaign launches. Legal/compliance teams can trust the system.
Why These Challenges Matter
Competitors punt on these:
- They offer generic Spanish TTS (not regional)
- They provide slow previews (wait and see)
- They require manual data uploads (no live integration)
- They don't handle multi-format delivery well
- They treat compliance as the client's problem
We solve them: Because our clients can't afford to ignore them. A Chilean AFP with 1M+ members needs Chilean Spanish. A bank launching a credit card campaign needs real-time balances. Insurance companies need compliance built-in.
These aren't just features. They're the difference between "cool demo" and "production at scale."
Tech Stack Summary
| Layer | Technology | Why |
|---|---|---|
| Client Rendering | Vite + React | Fast dev, instant previews, WebGL support |
| API Services | .NET 8 microservices | Performance, security, enterprise-grade |
| Website | Next.js 14 | SEO, static generation, edge deployment |
| Cloud Platform | AWS | LATAM presence, compliance certifications, mature services |
| Video Processing | MediaConvert + Lambda | Scalable, cost-effective, format flexibility |
| Data Storage | S3 + RDS (PostgreSQL) | Durability, encryption, compliance-ready |
| Caching | Redis (ElastiCache) | Speed, real-time preview support |
| CDN | CloudFront | Low latency across LATAM |
| Security | KMS + Secrets Manager | Encryption, credential rotation |
| Monitoring | CloudWatch + X-Ray | Observability, debugging, compliance logging |
What's Next
These challenges aren't solved on day one. They're our technical mission — the hard problems that make Percus worth building.
As we grow:
- Open-source parts of the TTS pipeline (share learnings)
- Publish benchmarks (storyboard generation times, rendering throughput)
- Build case studies (how we solved X for client Y)
- Attract engineering talent who want to solve hard problems
We're not just another video platform. We're solving LATAM financial services video personalization at scale.