class AathithyaArasu:
def __init__(self):
self.name = "Aathithya Arasu S"
self.location = "Chennai, India"
self.degree = "B.Tech Information Technology (Expected May 2027)"
self.college = "St. Joseph's College of Engineering"
self.cgpa = 8.20
self.stack = [
"PyTorch", "LangChain", "llama.cpp", "Ollama",
"FastAPI", "Qdrant", "FAISS", "Docker",
"AWS SageMaker", "OpenCV", "EfficientNetB7"
]
self.currently_learning = [
"Advanced Agentic RAG Architectures",
"Graph RAG & Knowledge Graph Retrieval",
"LLM Fine-tuning on Constrained Hardware",
"Java DSA for Interview Readiness"
]
self.fun_fact = (
"Built a production RAG pipeline during internship "
"that got adopted by the internal team in week one."
)
def motto(self) -> str:
return "Ship systems that work on real hardware, not benchmark rigs."
me = AathithyaArasu()Languages
ML / DL / CV
RAG / LLM
Cloud & DevOps
Databases & Retrieval
Tools
Ozis Technology — Software Engineering Intern | Feb 2026 – Mar 2026 | Madurai, India
PythonRAGGraph RAGLangChainVector RetrievalFastAPI
- Designed and deployed an end-to-end RAG pipeline adopted by the internal team for production document retrieval within the internship window.
- Benchmarked Naive RAG vs. Graph RAG across document sets, evaluating retrieval accuracy, latency, and scalability under real load conditions.
- Delivered structured performance comparisons that directly informed the team's architecture decision for their document Q&A product.
INFINEX Corporation Private Limited — Machine Learning Intern | Dec 2025 – Jan 2026 | Chennai, India
PythonNLTKScikit-learnTopic ModelingSentiment AnalysisLLM
- Built NLP pipelines for sentiment analysis and topic modeling on 10,000+ research documents, improving classification efficiency by 25%.
- Developed "NLP-based Sentiment Analysis and Topic Modeling for Research Integrity" — a full pipeline combining NLTK, Scikit-learn, and LLM-assisted extraction.
- Automated insight extraction from unstructured data, reducing manual analysis effort across large-scale document corpora.
| Project | Stack | Highlights |
|---|---|---|
| Multimodal RAG System V2 | FastAPI · Qdrant · BGE · CLIP ViT-B/32 · BM25 · llama.cpp · Docker | 4-stage hybrid retrieval (Vector + BM25 → RRF k=60 → reranker @ 0.15); 4-signal confidence scoring (45/25/15/15%); Qwen2.5-1.5B Q4_K_M running at ~2.5 GB on 4 GB VRAM |
| Multimodal Document Tampering Detection ⭐ 29 | EfficientNetB7 · ELA · Grad-CAM · EasyOCR · MC Dropout · AWS SageMaker | 6-signal cross-modal fusion: Grad-CAM (0.25) + Visual-OCR IoU (0.30) + OCR-visual conflict (0.15) + OCR confidence penalty (0.10) + MC Dropout uncertainty (0.10) + spatial density agreement (0.10) → tiered LOW/MEDIUM/HIGH risk; deployed on SageMaker real-time inference API |
| Confusion Matrix Debugger | Python · Streamlit · LangChain · FAISS · Groq API | LangChain + FAISS RAG backend for contextual error explanations; automated class-imbalance and mislabeling detection reducing manual evaluation overhead |
| Achievement | Details | |
|---|---|---|
| 🥇 | Cognizant Technoverse Hackathon 2026 — Finalist | National-level CTS hackathon; Personalised Banking theme; built Mu AI Finance Agent (bank statement ingestion, transaction categorisation, GST compliance, what-if simulation, Ollama Qwen AI chat) |
| 🥈 | Makethon 3.0 — Finalist | Feb 2026 |
| 🎓 | SIH 2026 — Top 50 College Team | Confirmed slot representing St. Joseph's College of Engineering |
| 🏅 | Python for Data Science — Silver Medal (Top 5%) | NPTEL / IIT Madras · Jul–Aug 2025 |
| ☁️ | OCI 2025 Generative AI Professional | Oracle University · Jul 2025 |
| 📜 | Introduction to NLP | Infosys Springboard · May 2025 |
| 📜 | AI Fundamentals | IBM SkillsBuild · Feb 2025 |
| Degree | Institution | Expected | Score |
|---|---|---|---|
| B.Tech — Information Technology | St. Joseph's College of Engineering, Chennai (Anna University) | May 2027 | 8.20 / 10 CGPA |
🧩 Advanced RAG → Graph RAG · HyDE · RAPTOR · Agentic RAG loops
🔧 LLM Deployment → GGUF quantisation · llama.cpp optimisation · speculative decoding
🤖 Agentic Systems → Tool-use · Multi-agent orchestration · ReAct / LATS patterns
☕ Java DSA → OOP deep-dive · Collections · Interview-pattern problem sets
🧠 CV Architectures → ConvNeXt · SAM · DINO · zero-shot classification