Hello, I'mSandeep!
I’m passionate about building end-to-end AI pipelines that solve real-world problems through machine learning and software engineering.
About Me
Sandeep Sahu
Aspiring AI/ML Engineer
I'm a recent graduate passionate about building intelligent systems that solve real-world problems. My interests lie in computer vision, NLP, and full-stack AI development, where I combine modern machine learning with strong software engineering principles to create scalable and impactful digital experiences. I enjoy transforming complex ideas into elegant, user-focused solutions powered by intelligent technology.
Education
B.E. in Computer Science & Engineering (AI/ML)
Lokmanya Tilak College of Engineering
CGPA: 7.95/10 · Sep 2023 — May 2026
Diploma in Automation and Robotics
Vivekanand Education Society Polytechnic
Grade: 81.33% · Jan 2021 — Jun 2023
Achievements
- ◆Finalist — CodeSummit@Sikkim'50 National Hackathon (May 2025)
- ◆MedBot — LTCoE Innovation Showcase (Apr 2025)
Certifications
- ◆NIIT Foundation — Cyber Smart AI Prime Internship (Cisco CSR / DSCI)
- ◆Lloyds Banking Group — Data Science Job Simulation (Forage)
- ◆JPMorgan Chase — Quantitative Research Virtual Experience (Forage)
- ◆ISRO — Bharatiya Antariksh Hackathon
- ◆Machine Learning & Data Science Masterclass — Udemy
- ◆Complete Python Bootcamp — Udemy
Technologies
Programming
Python
C
R
SQL
HTML/CSS
BashFrameworks & Libraries
Django
FastAPI
OpenCV
PyTorch
TensorFlow
Keras
Scikit-learn
Pandas
NumPy
MatplotlibSeabornAI/ML
Computer Vision
NLPBiLSTMAttention Mechanisms
RAG Pipelines
Semantic Search
LLM IntegrationDatabase
MySQL
Firebase
ChromaDB
pgvectorFirebase Realtime DatabaseTools
Git/GitHub
Jupyter Notebook
Google Colab
Cursor
Kaggle
Postman
Hugging Face
Android StudioDaphneMy Projects
Happy Club
AI Sign Language Translation
A person signs in front of a webcam; the app detects body pose, classifies the gesture, and outputs readable English (via BiLSTM + Gemini). In reverse, typing English produces sign output. Built for classrooms and daily use where a human interpreter is not always available—targeting low latency (~30 fps) on the web.
- ✦91.2% Top-1 accuracy across 79 ISL word classes
- ✦~30 fps streaming with 38ms average frame latency
- ✦Bidirectional pipeline: vision → BiLSTM → Gemini gloss translation
SatSR
Dual Image Super Resolution
Built for Bharatiya Antariksh Hackathon 2025 (Problem Statement PS-12): fuses two 128×128 low-resolution optical frames into one 384×384 super-resolved scene using a dual-branch SRCNN pipeline trained on 594 ESA PROBA-V scenes. Includes blind quality evaluation (PSNR, SSIM, NIQE) so outputs can be scored without a ground-truth reference at inference time.
- ✦PSNR 41.93 dB · SSIM 0.9633 · NIQE 11.77 on held-out evaluation
- ✦Blind-reference Spearman correlation 0.82 for quality ranking
- ✦Dual-branch SRCNN: 2× LR inputs → single HR optical output (3× upscale)
Legal AI
Legal Document Retrieval & Virtual Legal Assistant
Indexes 164K chunks from 33 legal sources (acts, codes, judgments). You ask a question in plain Hindi or English; the system retrieves the most relevant passages with semantic search + reranking, then Gemini answers using only that context. Two modes: Corpus Search (find a specific law or case) and Personal Assistant (describe your situation and get guided, classified legal help).
- ✦164K indexed chunks from 33 legal corpora
- ✦93.3% accuracy on 6-class legal problem classification
- ✦Dual-mode UX: corpus search + personal legal assistant
PneumoScan
Pneumonia Detection from Chest X-Rays
Upload a chest X-ray in the Streamlit app. A fine-tuned ResNet50 labels it Normal or Pneumonia. Grad-CAM overlays highlight which lung regions drove the prediction so clinicians can sanity-check the AI. Useful as a research/demo tool for explainable medical imaging, not a replacement for radiologists.
- ✦0.98 AUC-ROC and 97.7% precision on held-out data
- ✦Two-phase fine-tuning with optimized decision threshold
- ✦Streamlit deployment with live upload and lung heatmaps
Medbot
AI-Based Medical Assistant Chatbot
Android chatbot: the user describes symptoms in conversation; an NLP engine maps patterns to possible conditions and suggested next steps (triage-style guidance, not a diagnosis). Firebase handles login, chat history, and an admin dashboard for monitoring usage. Showcased at LTCoE Innovation Showcase 2025.
Jul 2022 – Apr 2023
Smart Blind Stick
GPS & GSM Navigation Aid
A smart walking stick with ultrasonic sensors for obstacles ahead and a rain sensor for wet surfaces. Immediate buzzer/vibration warns the user. On a critical hazard, a SIM800L module texts registered contacts with GPS coordinates. Built as embedded C firmware on real hardware—not only a software demo.
Professional Journey
Intern
Questcont Technology · Mumbai
- —Performed precision soldering and PCB assembly of resistors, capacitors, transistors, and diodes on warp detectors.
- —Conducted component-level diagnostics using multimeters and LED-based sensors to test diode values.
- —Executed ferrule cutting, wire measurement, and sorting. 1000+ multi-length industrial wires used in instrumentation systems.
- —Developed and maintained structured hardware diagnostic logs using basic Python scripting for PCB testing, validation, and component-level analysis.