🚀 Open to Software Engineering Internship Opportunities
📬 Connect with me on LinkedIn
- LLM Systems (RAG, Vector Databases, Prompt Engineering)
- Backend Engineering (FastAPI, REST APIs, Microservices)
- Machine Learning (NLP, Computer Vision)
- System Design & Scalable Architectures
- Cloud & DevOps (Docker, AWS, GCP)
Building scalable backend systems and production-ready AI applications that deliver real-world impact.
I’m a Computer Science undergraduate (2026) focused on backend engineering, system design, and applied AI/LLM systems. My work combines software engineering with machine learning to build reliable, scalable, and production-ready applications.
During internships at NIT Rourkela and Infosys Springboard, I developed and deployed real-world AI systems including:
- Face recognition pipeline with 96% accuracy using PyTorch and StyleGAN-based augmentation
- Real-time sign language recognition system achieving 95% accuracy
- Optimized inference and evaluation pipelines for scalable deployment
My work also includes research in edge AI and secure computer vision systems:
- 📄 Published IEEE research on deepfake-aware edge AI authentication
- Built a unified Raspberry Pi-based pipeline integrating YOLOv5 for face detection, FaceNet for recognition, and EfficientNet-B4 for deepfake detection
- Evaluated on benchmarks including WIDER Face, LFW, CelebA, and FaceForensics++
- Achieved real-time inference at ~15 FPS with 0.96 AUC for deepfake detection under challenging real-world conditions
- Optimized for secure, low-latency identity verification in edge AI, IoT, and surveillance environments
I also build LLM-powered systems such as:
- ⚖️ RAG-based legal document analysis platform processing 1K+ documents/day
- Reduced manual review time by 60% using retrieval and transformer-based pipelines
Tech interests:
Backend Systems • Distributed Systems • LLMs • RAG • System Design • Computer Vision • Cloud & DevOps
Currently seeking Software Engineering Internship opportunities (Backend & AI Focus) to build scalable systems and intelligent AI-powered products.
📊 Impact:
- Processes 1K+ documents/day
- Achieves 92% clause extraction accuracy
- Reduces legal review time by 60%
 |
 |
|---|---|
 |
 |
- 📄 Multi-format uploads (PDF, Word, Images)
- 🤖 AI-powered legal Q&A using NLP models
- 🧠 Clause paraphrasing with T5-based models
- ⚖️ Risk & sentiment analysis of contract clauses
- 🔐 Secure OTP-based authentication
- ⏳ Configurable session timer with automatic logout (10-minute default) for secure access control
- 📜 Session history tracking to review previous queries and responses
- 📥 One-click download of complete session reports as PDF
- 🛡️ Your document is 100% secure — no user data or uploaded documents are stored
- ⚡ Modern dashboard with Next.js & React
- 🐳 Dockerized deployment using Docker Compose
🔗 Repository: https://github.com/Soumitra1312/DocuLix
📊 Impact:
- Achieves 92% recognition accuracy
- Automates attendance, reducing manual effort and errors
 |
 |
|---|---|
 |
 |
- 📸 Automated image capture — captures 10 images per user via webcam
- 🧠 Face recognition system using OpenCV, dlib, and face_recognition
- 🗂️ MongoDB-backed storage for users, attendance, and schedules
- 👨🏫 Faculty dashboard with time slots, sections, and student lists
- 📊 Attendance analytics including history and per-class statistics
- ⚡ End-to-end automation eliminating manual attendance processes
🔗 Repository: https://github.com/Soumitra1312/Face-Recognition
📊 Impact:
- Delivers smooth real-time gameplay with optimized rendering and AI behavior
 |
 |
|---|---|
 |
 |
- ⚔️ Enemy AI with behavior trees for chasing, flanking, and attacking players
- 🔫 Realistic shooting mechanics with aiming, firing, and hit feedback
- 💥 Gun recoil & audio effects for immersive combat experience
- 🏃 Fluid player movement including walking, sprinting, jumping, and dodging
- 🧠 Responsive control system tuned for fast-paced third-person gameplay
- 🎮 Action-focused combat loop rewarding movement and situational awareness
🔗 Repository: https://github.com/Soumitra1312/Shadow-Fire
| Publication | Venue | Research Area |
|---|---|---|
| Deepfake-Aware Face Authentication for Edge Devices Using a Unified Raspberry Pi Pipeline | IEEE AIEI 2026 | Edge AI, Computer Vision, Deepfake Detection |
- Published in IEEE AIEI 2026
- Built a complete edge AI pipeline integrating YOLOv5, FaceNet, and EfficientNet-B4
- Achieved real-time face detection at ~15 FPS on Raspberry Pi
- Reached 0.96 AUC for deepfake detection using FaceForensics++ and real-world camera inputs
- Optimized for secure and low-latency identity verification on resource-constrained edge devices
- Evaluated across challenging real-world conditions including masks, sunglasses, occlusion, and varying lighting conditions
📄 Read the Full Paper Here : Link
“Programs must be written for people to read,
and only incidentally for machines to execute.”
— Harold Abelson