U Saran
Raj.
Final-year B.Tech AI & Data Science student — studying not just how AI works, but where it breaks, why it fails, and how to make it trustworthy.
120 days inside AI's failure modes — completed.
A completed independent research challenge — not building more models, but understanding where they break, why they fail, and how to make them accountable. The reports below are its output.
"The gap between a working model and a production-ready AI system is far larger than most tutorials acknowledge.
Data Science
Data robustness: data decay, missingness patterns, and proxy variable risks in real-world datasets.
Machine Learning
Model fragility: generalization gaps, distribution shift, and failure under edge cases.
Deep Learning
AI systems: neuron behaviour analysis and human-in-the-loop dynamics in deployed systems.
GenAI & Agents
Agentic safety: RAG failure modes, hallucination patterns, and autonomous agent safety constraints.
Systems built with intent.
Five projects, one shared question — when does this break, and can we prove it won't?
"Most people build AI to impress. I build AI to interrogate. A model that scores 99% but can't explain itself is not intelligence — it's a pattern with confidence. The most important question in AI is not "does it work?" but "when does it break, who does it hurt, and can we see why?""
Robustness is not a feature. Fairness is not optional. Explainability is the baseline.
How I think.
Every AI system I build follows this framework — not a rigid checklist, but a way of thinking that keeps the real problem in focus.
Where I've shipped.
Full-Stack Developer Intern
- →Built 8 full-stack applications within the program using React.js, Node.js, Express.js, and MongoDB.
- →Designed REST APIs, tested at around 200 requests per minute in program workloads.
- →Set up CI/CD pipelines and automated tests for the program projects.
- →Collaborated in agile teams using Git-based version control and structured code reviews.
Python Programming Intern
- →Completed 4 structured Python assignments covering core language constructs.
- →Wrote and tested reusable Python scripts, focusing on correctness and clear execution flow.
- →Participated in regular progress reviews and incorporated supervisor feedback.
What I reach for.
Full stack at a glance — no scroll, no wait.
AI / Machine Learning
- Machine Learning
- Deep Learning
- Computer Vision
- NLP
- TensorFlow
- PyTorch
- Sentence Transformers
- FAISS
AI Robustness & Ethics
- SHAP Explainability
- Bias Detection
- Fairness Metrics
- Model Robustness
- RAG Failure Analysis
- Responsible AI
Cloud & Backend
- AWS EC2
- DynamoDB
- S3
- Amazon Bedrock
- FastAPI
- REST APIs
- MongoDB
- PostgreSQL
- Redis
- Neo4j
Languages & Tools
- Python
- JavaScript
- React.js
- Node.js
- Docker
- Git
- CI/CD
- Streamlit
Currently learning.
Agentic AI & RAG Systems
RAG failure modes, autonomous agent safety, hallucination patterns in production.
Advanced Mathematics for ML
Linear algebra, probability theory, and optimisation for deep learning foundations.
Korean Language
Vocabulary and reading comprehension, working toward TOPIK certification.
Cloud Architecture (AWS)
Scalable inference, serverless pipelines, cost optimisation for AI workloads.
Open to opportunities.
I'm actively seeking roles where AI is taken seriously — not just shipped fast. If you're working on systems that need to be robust, explainable, or fair, let's talk.
AI Research Intern
Open ↗Model robustness, failure analysis, safety evaluation, responsible AI research.
ML Engineer Intern
Open ↗Building and evaluating production-ready ML pipelines with real constraints.
Full-Stack AI Developer
Open ↗End-to-end AI applications from model training to cloud deployment.
Open Source Contributor
Open ↗Contributing to AI fairness, robustness, and explainability tooling.