Building ViHi: A Video-First Gig Platform
At Neosenth, we're building ViHi: a video-first gig platform where you post a short video of your problem, get AI-matched professionals, compare bids, and hire—or earn as a pro. This post is about why we chose this direction and how we're building it with scalable systems, video, and AI.
Why video-first?
Text and photos don't capture “I need this fixed by Thursday” or “here’s exactly what’s broken.” Video does. We wanted a format that:
- Reduces back-and-forth and miscommunication
- Lets pros assess the job quickly and bid accurately
- Gives clients a clearer picture of who they’re hiring
So the core interaction is post a short video → get matched → compare and hire. That required rethinking both the product and the stack.
What we're building
ViHi isn’t just another job board. We’re putting video and AI at the center.
For people who need help
- Record a short video of the problem (e.g. a broken fixture, a design ask).
- Submit it on the app; our pipeline processes and indexes it.
- Get matched with relevant pros via AI (skills, context, and availability).
- Compare bids and hire; payment and scheduling live in the platform.
For professionals
- Create a pro profile and show your work via short clips.
- Receive matched requests instead of hunting for posts.
- Bid on jobs that fit; earn through a fair subscription model instead of per-job fees.
Horizontal rule
A clean separation before we get into the stack:
Below, the tech choices that make this possible.
Architecture and tech stack
My background is in fleet tracking, real-time analytics, and AI-powered products—systems that handle millions of records per minute. For ViHi we needed:
- Low-latency APIs and real-time updates
- Reliable ingestion for video metadata and events
- ML pipelines for matching and search
- Cross-platform clients (mobile and web) with a single codebase where it made sense
So we chose a stack that plays to those needs.
Backend and data
| Layer | Technology | Role | | ------------ | ------------ | ----------------------------------------- | | API & services | Go, ConnectRPC / gRPC | Fast, type-safe APIs and service-to-service calls | | Data store | PostgreSQL | Users, jobs, bids, subscriptions | | Cache & sessions | Redis | Sessions, rate limits, real-time state | | Events | Kafka | Video upload events, matching jobs, notifications |
We use ConnectRPC (and gRPC where it fits) so that mobile and web clients get a consistent, typed contract and streaming where we need it.
Client and ML
- Flutter and Dart for iOS, Android, and web. One codebase for core flows and UI.
- PyTorch and TensorFlow for matching and any on-server ML (e.g. embeddings, ranking).
- OpenCV (and related tooling) for video thumbnails, basic analysis, and metadata extraction.
A small example of the kind of structure we use on the backend for job matching:
type MatchRequest struct {
JobID string `json:"job_id"`
VideoRef string `json:"video_ref"`
Categories []string `json:"categories"`
Limit int `json:"limit"`
}
func (s *MatchingService) FindPros(ctx context.Context, req *MatchRequest) ([]Pro, error) {
// Embed job + video metadata, query vector store, rank by relevance & availability
// Return top N pros for bidding
}
Inline code like MatchRequest and FindPros lives in a larger Go codebase that stays explicit and testable.
Blockquote
Good systems are boring in the right places: consistent APIs, clear data models, and observable pipelines. The magic is in how video and AI improve the experience, not in reinventing the wheel everywhere.
We keep the “boring” parts solid so we can iterate on matching, UX, and fairness.
Fair model and launch
We’re launching with a subscription model so pros aren’t nickel-and-dimed per gig. Clients get a clear path from video → match → hire without endless scrolling. We’re iterating on both sides of the marketplace.
If you’re interested in backing, partnering, or just following along, I’d love to connect. You can reach out via codesark.me or Neosenth.
Building with impact — Savinay Kumar, founder of Neosenth.