Semantix

AI-powered semantic search that understands user intent, not just keywords.
System-level design with real-time indexing and vector database integration.
Clean, modular architecture built by first-year students.
Robust backend with health monitoring and controlled indexing.

Inspiration

As first-year college students, we were curious about how modern search engines understand meaning, not just keywords. While balancing classes, exams, and assignments, we wanted to build something beyond the syllabus — a system that truly thinks from the user’s point of view. This curiosity pushed us to explore semantic search and AI-driven retrieval.

What it does

Semantix is an AI-powered semantic search engine.
Instead of matching keywords, it understands the intent behind a user’s query and returns results based on meaning. Users can type naturally, and the system responds with the most relevant information.

How we built it

We split the work based on our strengths:

Front-end: Designed a clean, simple interface focused on user experience.
Back-end: Built APIs and logic to connect the system end-to-end.
Search & Optimization: Implemented embeddings, vector similarity, indexing, and performance optimizations.

We integrated everything using a modular architecture so each part worked smoothly with the others.

Challenges we ran into

Managing time as first-year students with heavy academics.
Debugging environment and import issues on Windows.
Handling search requests before indexing was complete.
Reducing cold-start latency and improving performance.

Each challenge taught us patience, teamwork, and problem-solving.

Accomplishments that we're proud of

Building a full semantic search system from scratch.
Designing a search engine that understands user intent.
Implementing optimizations like model warm-up and caching.
Successfully integrating frontend, backend, and search logic.

Most importantly, we’re proud that we built something real and usable as first-year students.

What we learned

AI is more than models — it’s about systems and user experience.
How embeddings and similarity search work in real applications.
The importance of clean architecture and collaboration.
Learning by building teaches more than theory alone.

What's next for Semantix

Adding support for PDFs and documents.
Improving ranking and search accuracy.
Deploying the system online for public use.
Expanding Semantix into a study and knowledge-search tool for students.

Semantix started as a learning project — and we’re excited to see how far it can go.

Built With

css
custom-vector-database
faiss
flask
flask-cors
git
html
javascript
numpy
python
pytorch
rest-apis
sentence-transformers

Submitted to

Build From Scratch - Season I

Created by

I worked on the search engine logic, optimization, and API integration for the project. My role focused on building the core semantic search system that understands user intent rather than relying on keyword matching.

I implemented text embedding generation, vector-based similarity search, and document indexing to ensure accurate and meaningful search results. I also optimized the system by adding controlled indexing, model warm-up, and caching strategies to improve performance and reduce response latency.

In addition, I integrated the search engine with the backend APIs, designed readiness checks using a health endpoint, and ensured the frontend could safely interact with the backend without querying before indexing was complete. My contribution helped make the system stable, efficient, and user-focused.

Salmon Angelo
Debugged and optimized application code to ensure stability and performance.
Implemented and supported machine learning components.
Managed and maintained GitHub repositories and version control workflows.
Handled API key management, security practices, and integration processes.
Created and edited videos for project demonstrations and presentations.

Srihari T
visnu mitran