Yashraj Shrivastava
I build systems that pull signal out of noise — inside encryption ciphers, financial markets, and machine learning models. Currently a B.Tech Computer Science student at Symbiosis Institute of Technology, Pune.
Where the interests meet
My work sits at the overlap of three fields that all, underneath, ask the same question: how do you separate a true pattern from randomness? In cryptography that's designing ciphers that look like pure noise to an attacker. In markets, it's filtering a real trading signal out of volatile, noisy price action. In machine learning, it's teaching a model to read faint emotional or behavioural cues buried in messy, multimodal data.
I've worked on all three — lightweight cipher design at DRDO ITR (Defence Research and Development Organisation, Integrated Test Range), market analysis at a finance firm, and independent projects spanning trading-bot risk management and affective AI.
Symbiosis Institute of Technology, Pune
New GreenField Public Academy
Recent roles
Summer Intern
- Designed highly secure, lightweight S-boxes (substitution boxes) for data encryption on low-power devices without compromising memory efficiency.
- Trained a Reinforcement Learning (RL) agent to navigate massive combinations, using semi-bent Boolean functions to guarantee perfectly balanced, unpredictable code.
- Tested grid architectures using 3-bit and 4-bit Cellular Automata neighborhoods to identify optimal spatial layouts for data-scrambling patterns.
- Discovered a lightweight configuration achieving a perfect nonlinearity score of 112, matching heavy-duty AES (Advanced Encryption Standard) industry benchmarks with far lower computational overhead.
Intern
- Gained hands-on exposure to financial markets, portfolio construction, and risk assessment.
- Analyzed equities, bonds, and mutual funds, and conducted small/mid-cap market evaluations.
Selected builds
AlgoSentinel
- Built a risk-management microservice using a PyTorch TCN (Temporal Convolutional Network) Autoencoder with 1D causal dilated convolutions to process 3D log-telemetry tensors.
- Engineered a low-latency ingestion pipeline via Redis Streams and a Pub/Sub emergency circuit that triggers KILL / FLATTEN commands against Binance and Zerodha APIs (Application Programming Interfaces) to prevent liquidation.
- Integrated SHAP (SHapley Additive exPlanations) as a real-time explainability layer to compute feature-attribution scores on anomalous telemetry windows.
JOY
- Engineered a multimodal virtual assistant using affective computing and Natural Language Processing (NLP) to deliver personalized, context-aware therapeutic interactions.
- Architected conversational flows integrating real-time sentiment analysis and multimodal data pipelines to dynamically adjust responses to nuanced emotional states.
- Implemented emotion-recognition models using Python, NLTK (Natural Language Toolkit), and Hugging Face Transformers for affective-state classification.
Tools & concepts
Programming
Data Science
AI & Machine Learning
Frameworks & Libraries
Web & Cloud
Databases
Beyond the code
EPIC Cell — SIT Pune
Coordinated technical events and led sponsorship outreach, supporting large-scale events on budget.
StartupCon 4.0
Directed event budgets and financial planning, engaging sponsors and stakeholders through targeted proposals.
Credentials
- AWS (Amazon Web Services) Cloud Practitioner Essentials
- Complete Data Science, Deep Learning (DL), Machine Learning (ML) & NLP Bootcamp — Krish Naik (Udemy)