booting portfolio kernel...
[--------------------] 0%
installing packages / compiling modules / linking ui runtime
SOFTWARE
DEVELOPER
I build digital products that refuse to be boring.
Python • JavaScript • Java • C++ • Node
AVATAR.JPGWho am I?
I am a software engineer driven by the challenge of bridging the gap between cutting-edge AI research and scalable production environments. Whether it is fine-tuning LLMs for domain-specific tasks or architecting real-time collaboration platforms, I focus on building systems that are as technically robust as they are user-centric.
</> Production-Grade AI: Experienced in developing RAG and fine-tuning LLMs
</> Full-Stack Mastery: Proficient in creating seamless and robust architectures.
</> Algorithmic Excellence: A Knight on LeetCode (Top 9% globally) with a 1st place finish at Coders' FaceOff.
TECH_DOMAINS
/// domain expertise ///
> Frontend_
- Redux
- Tailwind CSS
> Backend_
> Databases_
> AI & Data_
- LLM Fine-tuning
- Data Analysis
- Pytorch, TensorFlow, Scikit-learn
> Architecture & Design_
- Algorithmic Problem Solving
- Microservices
- Pub/Sub Systems
- Real-time Collaboration Systems
My Journey
Journey Timeline
Backend Developer Intern @ Naalanda AI
Feb 2026 - Present
Bangalore
- Developing a scalable RAG system for an EdTech platform, integrating vector databases and fine-tuning LLMs to deliver personalized learning experiences.
Software Engineer Intern @ SGF-Tech
Dec 2025 - Present
Mumbai
- Architecting a Headless E-commerce platform using Medusa.js and Mercurius, managing complex product logic within a scalable Monorepo.
Full Stack Developer @ Vanivert AI
July 2025 - Dec 2025
Una
- Architected a real-time conversational Voice Agent integrating STT, LLMs, and TTS engines to achieve human-like latency and intonation.
AI Research Intern @ Sleath Startup
Feb 2025 - June 2025
Lucknow
- Analyzed large-scale marketing datasets using Pandas and NumPy to identify key influencer engagement trends and performance metrics.
Data Analyst @ Fluencer Digital
May 2024 - July 2024
Ranchi
- Analyzed large-scale marketing datasets using Pandas and NumPy to identify key influencer engagement trends and performance metrics.

CODING_STATS
LIVE
GITHUB
roshan3101
Midnight Coder
Commits
Matrix_Output
LEETCODE
roshanSahu3101
Problem solver
Ranking
Selected Works

CollabTask
Real-time enterprise collaboration platform with multi-tenant workspace architecture.
▸Engineered a high-performance event infrastructure using WebSockets, delivering sub-100ms real-time updates for team activity feeds.
▸Resolved write conflicts in multi-user environments by implementing Optimistic Locking, ensuring data integrity without compromising throughput.
▸Achieved end-to-end type safety by synchronizing Zod and Pydantic schemas, accelerating developer velocity by ~40%.

IntelliScrape
No-code web automation SaaS for designing and executing visual scraping workflows.
▸Developed an interactive visual builder that parses JSON-based Directed Acyclic Graphs (DAGs) to execute multi-step automation tasks.
▸Designed a consumption-based credit system to track resource usage and prevent API abuse.
▸Secured sensitive user credentials using AES-256 encryption at rest for safe injection into automation workflows.

Slimnastics
Comprehensive Gym Management & E-Commerce SaaS solution.
▸Architected a secure RBAC system and JWT httpOnly cookie flow to isolate data across Admin, Trainer, and Member roles.
▸Integrated Razorpay with server-side validation for atomic transactions and order history management.
▸Optimized frontend performance by ~35% using Redux Toolkit for global state caching and Vite for optimized bundling.

LLaMA 3.1 Fine-Tuning Pipeline
Production-grade modular framework for training and deploying domain-specific Small Language Models (SLMs).
▸Developed a QLoRA fine-tuning pipeline for 8B+ parameter models, utilizing 4-bit quantization to reduce training costs by ~60%.
▸Optimized inference for edge devices by quantizing models to INT4 GGUF format, enabling real-time CPU execution with <1GB RAM.
▸Built an OpenAI-compatible API layer to serve local models as drop-in replacements for cloud-based LLMs.
Data Capsule
Education & Achievements
Education
Academic LedgerIndian Institute of Information Technology, Una
B.Tech in Electronics and Communication Engineering
Delhi Public School, Ranchi
Intermediate (Class 12)
Oxford Public School, Ranchi
High School (Class 10)
Achievements
Signal BoostKnight on LeetCode
Achieved 'Knight' status on LeetCode by solving 600+ problems, ranking in the top 7.9% of competitive programmers globally.
1st Position at Coders' FaceOff
Awarded 1st Position at Coders' FaceOff (IIIT Una, 2024), demonstrating proficiency in advanced data structures and algorithms.
3rd Position at Electrothon
Secured 3rd Position at Electrothon (NIT Hamirpur, 2024), developing a high-impact solution within a 48-hour sprint.
Hackathon Finalist
Selected as a finalist and participant in premier hackathons including HackTU 5.0 (Thapar) and HackTheHills (IIIT Una).
USER_REPORTS.txt
Architected a production-grade Computer Vision Parking Slot Detector now managing real-time occupancy across 30+ commercial facilities with exceptional reliability.
Developed a fine-tuned LLM blending career guidance with empathetic support, significantly improving user engagement and automated assistance quality.
Built a low-latency AI Voice Agent for virtual meetings, delivering natural interactions and optimized real-time communication performance.
Architected a production-grade Computer Vision Parking Slot Detector now managing real-time occupancy across 30+ commercial facilities with exceptional reliability.
Developed a fine-tuned LLM blending career guidance with empathetic support, significantly improving user engagement and automated assistance quality.
Built a low-latency AI Voice Agent for virtual meetings, delivering natural interactions and optimized real-time communication performance.
Let's
Talk
Code.
I am currently available for freelance work and open to full-time opportunities.