Kanishk Ranjan Barman Logo
krbarman.simplyfyy.com

Projects Portfolio

An in-depth review of the system designs, challenges, and core solutions.

AI Powered HRMS App

Feb 2026 – Present

Automated candidate hiring CRM with LLM tool calling

Live Demo

A premium full-stack HRMS where HR teams create job forms, manage applicants, and onboard candidates via an interactive CRM interface. Integrates an Ollama LLM with tool-calling capabilities to autonomously send emails and update statuses through chat.

Key Highlights

  • •Built a full-stack HRMS with candidate CRM and Ollama LLM with tool-calling capabilities that autonomously sends emails and updates candidate statuses via chat.
  • •Implemented semantic candidate search using pgvector and Voyage AI embeddings to rank and match applicants with high relevance.
  • •Secured application with Google OAuth2 and JWT-based role-based access control, storing resume uploads in AWS S3 and sending transactional emails through Brevo.
  • •Deployed on AWS EC2 with Cloudflare routing and Vercel for frontend, with automated CI/CD using cron jobs and shell scripts.

Core Capabilities

Candidate CRM Interface
AI Assistant Chat
Semantic Search via pgvector
Autonomous Email Automation
Secure Google OAuth2 & JWT

Technology Stack

NextJSSpring BootSpring AIpgvectorVoyage AIOllama LLMAWS S3AWS EC2Google OAuthJWTTailwind CSSBrevo

Challenges Faced

Configuring Ollama's local LLM tool calling on EC2 with stable latency. Resolving this required implementing model parameter optimizations and a robust queue system for background processing of candidate emails and CRM status updates.

Key Learnings

Deepened knowledge in vector databases (pgvector), building resilient AI agent loops using Spring AI, and managing AWS EC2 scale-ups for hosting local models.

System Architecture Flow

Next.js
Next.js ClientFrontend / UI
Spring Boot
Spring Boot APIBackend Gateway
PostgreSQL
Postgres + pgvectorSemantic Storage
Ollama (LLM)Spring AI Agents
Data flow pipeline: User interactions map to vectorized DB embeddings, routing queries directly through Spring AI to locally hosted model pools for tool calling.

Real-Time Chat Application

Feb 2025 – March 2025

Highly concurrent multiroom chat engine

View Source

A highly responsive real-time chat application with a clean UI, enabling instant room creation, room joining, and concurrent user messaging. Includes full message history persistence.

Key Highlights

  • •Developed a real-time chat application allowing users to instantly create and join chat rooms.
  • •Implemented full chat history persistence with timestamps so new participants can follow previous conversations seamlessly.
  • •Containerized with Docker for easy local deployment and multi-environment setup.

Core Capabilities

Instant Chat Room Creation
WebSocket Message Streaming
Message History & Persistence
Dockerized Deployment Setup

Technology Stack

JavaSpring BootWebSocketsReactJSDockerTailwind CSSJavaScriptMavenLinuxGit

Challenges Faced

Handling concurrent connection spikes and message delivery order consistency. Addressed by designing a custom message buffer and thread-safe session handlers in Spring Boot.

Key Learnings

Mastered WebSocket protocols, state sync between client/server, and local multi-container orchestration with Docker.

WebSocket Connection Pipeline

Step 01
Handshake

HTTP request upgraded to WS protocol using SockJS fallback support.

Step 02
STOMP Broker

Subscribes client to specific chat-room routing endpoints (e.g. /topic/room).

Step 03
Message Sync

Bidirectional message streaming with DB persistence for history retrieval.