Vibe-Detecting AI Agent Build
Budget: $15 – $25 USD
I’m building Loveable.dev and need an AI expert to create an autonomous agent whose core talent is reading a user’s “vibe level.” In other words, the model should ingest short text (think chats, comments or posts) and instantly return a numerical or categorical score that captures the overall emotional tone, friendliness and positivity of the input.
Here’s how I picture it working:
• A simple REST or GraphQL endpoint receives the text.
• The agent analyses sentiment, contextual nuance and subtle cues, then outputs the vibe score plus a one-sentence rationale.
• Everything runs in real time so the response feels instantaneous inside the product.
You are free to combine large-language-model techniques, custom embeddings, emotion classification models or reinforcement learning with human feedback—whatever achieves accurate, explainable results. I’m open to Hugging Face Transformers, OpenAI, LangChain, or a small bespoke model if that makes more sense for speed and cost.
Deliverables
1. Fully trained model (or fine-tuned weights) with reproducible training pipeline
2. Clean, well-commented code for the inference microservice
3. Quick-start README plus a short Loom walk-through showing the agent in action
Acceptance criteria
• ≥ 90 % accuracy against an agreed-upon test set of labelled vibe levels
• Average latency < 700 ms per request on standard cloud hardware
• Easy to deploy to Docker/Kubernetes or a basic VPS
If you have prior work in sentiment analysis, emotion detection or conversational AI agents, I’d love to see it. Let’s make online interactions a little more loveable.
Here’s how I picture it working:
• A simple REST or GraphQL endpoint receives the text.
• The agent analyses sentiment, contextual nuance and subtle cues, then outputs the vibe score plus a one-sentence rationale.
• Everything runs in real time so the response feels instantaneous inside the product.
You are free to combine large-language-model techniques, custom embeddings, emotion classification models or reinforcement learning with human feedback—whatever achieves accurate, explainable results. I’m open to Hugging Face Transformers, OpenAI, LangChain, or a small bespoke model if that makes more sense for speed and cost.
Deliverables
1. Fully trained model (or fine-tuned weights) with reproducible training pipeline
2. Clean, well-commented code for the inference microservice
3. Quick-start README plus a short Loom walk-through showing the agent in action
Acceptance criteria
• ≥ 90 % accuracy against an agreed-upon test set of labelled vibe levels
• Average latency < 700 ms per request on standard cloud hardware
• Easy to deploy to Docker/Kubernetes or a basic VPS
If you have prior work in sentiment analysis, emotion detection or conversational AI agents, I’d love to see it. Let’s make online interactions a little more loveable.