By
@prthm
An AI-driven technical interview platform designed to conduct adaptive, structured candidate assessments. Using Large Language Models (LLMs) such as Llama 70B via Groq, Conlatus dynamically evaluates candidate responses against pre-configured rubrics, asking context-aware follow-up questions while maintaining auditable, deterministic scoring rules.
@prthm
on
Conlatus : AI screening Interview
Conlatus : AI-powered screening interviews
I built Conlatus, an AI-powered screening interview system designed to automate the repetitive first stage of candidate screening.
The Idea is :
Create interview -> Candidate enters -> AI conducts interview -> Responses are processed -> Recruiter reviews results.
The challenge ?
The hardest part wasn’t simply connecting an AI API. It was turning an LLM into an actual interview workflow.
I had to figure out how to manage interview state, connect the frontend and backend, process candidate responses, and structure the system so it could grow beyond a simple chatbot.
The stuff that I’m proud at…
It has a working backend, frontend interfaces, AI integration, environment configuration, documentation, and a reproducible local setup.
Test it..
You’ll need Python and a Groq API key (LLM Provider).
.jpeg)
git clone https://github.com/conlatus/conlatus.git
cd conlatus
python -m venv .venv
# Windows
.venv/Scripts/activate
# Linux / MacOS
source .venv/bin/activate
pip install -r requirements.txt
# Create .env from .env.example
cp .env .env.example
# Update your GROQ API : GROQ_API_KEY=_groq_api_key
# Start the server
uvicorn backend.main:app --reload
The app will be runnning on http://localhost:8000/.
And access admin dashboard on http://localhost:8000/admin/
note: Conlatus was mostly developer with the help of AI coding IDEs, especially Google Antigravity. I used it extensively for implementation, debugging, refactoring, architecture exploration and UI development.
What I did. Is that I directed the development process, made architectural decisions, reviewed the generated code, tested the application, and iterated on implementation.
I’ve actually created a user friendly website. You can find it at : https://conlatus.github.io/
@prthm
on
Conlatus : AI screening Interview
1h 7m 14s logged
Hey everyone, I just gave the Conlatus UI a massive brand makeover! :D
We waved goodbye to those generic glowing dark-mode vibes and said hello to a super sharp, minimal monochrome look.
The admin, candidate, and dashboard pages all got this sleek new treatment.
Out with the neon colors, and in with a gorgeous off-white and near-black palette.
We switched over to the Inter font so everything is wonderfully tight and readable.
I also squared off those bubbly rounded corners to a crisp 2px and ditched the heavy drop shadows for elegant, subtle borders.
Those old placeholder badges are finally out of here, replaced by our official SVG logo!
Oh, and the candidate chat is now completely stripped of colored bubbles so you can focus entirely on the actual conversation. ;)
The best part is that I pulled all of this off using just vanilla CSS and variables, without bringing in any heavy frameworks. B-)
I really hope you all love this clean new aesthetic as much as I do! <3
@prthm
on
Conlatus : AI screening Interview
30m 13s logged
As of this project sounds a bit of corporate. It isn’t..
I’m onboard this project thanks to my college’s “Micro Project”.
I’m going to show this project as a “Micro Project” in my college.
The status of the application so far is that only I have make it crawl (not even working as intended).
You can say that current status is the MVP of MVP.
I’ve only attached an LLM using the GROQ’s API (Llama’s 70b model).
And I’m planning to use a smaller model (maybe: Llama 8b). Because..
I found out that GROQ only gives 500k tokens per day. And mostly that becomes half if we use bigger model like 70b or GPT OSS 120b.
So, even though my application doesn’t need that much thinking power. I’ll be planning to use better context pipeline to feed it welll even in the small model.
I’ll be sharing my architecture of this project soon…
Happy hacking ;)
