TechShortsApp
A credibility-first short-form learning platform exploring how technical content can be ranked using behavioral evidence rather than popularity alone.
TechShortsApp is a research-driven short-form learning platform built around a difficult question: when truth and learning cannot be observed directly, what behavioral evidence can responsibly influence content ranking?
Active
The problem
Most short-form platforms optimize for engagement signals such as views, likes, completion, and retention. These signals may describe attention, but attention is not automatically evidence that content is correct, useful, or educational. TechShortsApp explores whether technical learning content can be ranked through a more cautious evidence model. The system does not claim to rank truth directly. Instead, it ranks confidence using uncertain behavioral traces.
- Truth is not directly observable from user behavior
- High engagement does not guarantee credibility
- The same behavioral signal may mean different things across contexts
- Ranking decisions must still be made despite incomplete evidence
What I built
I independently designed and developed the platform, including the public interface, authentication, video infrastructure, evidence collection, ranking logic, database models, moderation controls, documentation, deployment, and iterative versioning process.
- Next.js and TypeScript application architecture
- Node.js and MongoDB backend
- Video upload, storage, playback, and deletion workflows
- User-video evidence schema
- For You, Most Trusted, and New & Emerging feeds
- Bookmark, Helpful, Follow, replay, and watch-ratio signals
- Invite-only account system
- Administrative moderation and publishing tools
- Public technical documentation
The evidence model
The current model combines implicit and explicit behavioral signals while treating each signal as uncertain evidence rather than direct proof of quality. Ranking uses posterior estimates, confidence penalties, time decay, and domain-specific assumptions.
- Watch ratio records viewing behavior but remains ambiguous
- Replay may indicate interest, confusion, verification, or usefulness
- Bookmark represents an intention to revisit
- Helpful records explicit perceived usefulness
- Follow captures creator affinity rather than direct video correctness
- Domain half-lives account for differences in knowledge decay
How the system evolved
The project has passed through several versions. Important assumptions changed when implementation and research revealed that the original interpretations were too simple.
- Version 1.0 was deprecated
- The project moved from engagement-oriented ranking toward an epistemic evidence model
- The skip penalty was removed because early exit has multiple possible causes
- Helpful was gated by sufficient watch progress
- Creator affinity was separated conceptually from video usefulness
- Domain-specific time decay was introduced
- The system language changed from ranking truth to ranking confidence
Technical evidence
The project is publicly deployed and supported by working user interactions, technical documentation, research writing, version history, database schemas, ranking logic, and continued product iteration.
- Live application at techshortsapp.com
- Documentation at docs.techshortsapp.com
- Working video feeds and evidence interactions
- MongoDB user-video evidence storage
- Versioned ranking architecture
- Public research essays
- Deployment and infrastructure history
Engineering problems solved
Building the system required solving problems beyond the visible interface, including evidence consistency, storage behavior, ranking assumptions, concurrency, caching, and lifecycle management.
- Prevented duplicate evidence records caused by race conditions
- Fixed video deletion and Cloudflare R2 caching behavior
- Designed infinite scrolling feeds
- Separated draft assumptions from implemented ranking behavior
- Rebuilt deprecated versions instead of preserving weak foundations
- Developed confidence-oriented ranking under incomplete evidence
Current status
TechShortsApp remains an active software and research project. It currently functions as both a working platform and an experimental environment for investigating credibility, learning evidence, behavioral signals, and ranking under uncertainty.
- Current generation: Version 2.x
- Core evidence signals implemented
- Public deployment active
- an early invite-only group of users and technical videos
- Research writing ongoing
- Context-dependent signal interpretation remains an open question
- Future work includes knowledge-state modeling and external evaluation
What this work demonstrates
TechShortsApp is evidence of my ability to move from an abstract question to a deployed technical system, maintain it over time, and revise the model when the original assumptions fail.
- Full-stack software development
- Database and evidence-schema design
- Ranking-system architecture
- Product design and iteration
- Research synthesis
- Technical documentation
- Independent execution
- System debugging and maintenance
- Reasoning under uncertainty
- Willingness to replace weak assumptions
Current research questions
The project has generated deeper questions that now guide the next stage of research and development.
- Are behavioral signals inherently context-dependent?
- How should signal meaning change across technical domains?
- How does prior knowledge affect behavioral interpretation?
- Can creator trust and video usefulness be modeled separately?
- What evidence would show that the ranking system improves learning?
- When is the model sufficiently understood to justify broader deployment?