┌─────────────────────────────────────────────────────────────────────┐
│ ___ _ _ │
│ / __(_)_ ____ _| |___ _ _ __ __ _ _ _ │
│ \__ \ \ V / _` | / / || | ' \/ _` | '_| │
│ |___/_|\_/\__,_|_\_\\_,_|_|_|_\__,_|_| │
│ |
│ Decoding chaos into elegant equations. │
│ │
└─────────────────────────────────────────────────────────────────────┘
name: Sivakumar Ramakrishnan
role: Applied AI Engineer @ Ricoh USA
education: MS Data Science, CU Boulder (4.0 GPA)
location: Boulder, Colorado
motto: "Passionately curious. Relentlessly learning"- Building Agentic AI solutions for service documentation at Ricoh
- Architected RAG systems serving millions of users at Zoho
- Published models on HuggingFace (toxicity classifier, invoice parser)
- Kaggle competition winner (1st place - Goodreads prediction)
- Research in quantum computing, edge AI, and multimodal systems
AI & Machine Learning
| Domain | Technologies |
|---|---|
| Deep Learning | PyTorch, TensorFlow, JAX, Transformers, VLLM |
| NLP | RAG, RLHF/DPO, Seq2Seq, Attention, spaCy, Whisper |
| Computer Vision | CNNs, ViT, EfficientNet, YOLO, CLIP, OpenCV |
| Multimodal AI | PaLiGemma, TimeSformer, Cross-modal retrieval |
| MLOps | Weights & Biases, Ray, Streamlit, HuggingFace Hub |
Languages & Tools
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
~/projects/
├── zoho/
│ ├── rag-architecture Multi-modal RAG system, millions of users
│ ├── mamba-transformer Mamba SSM + Transformer hybrid architecture
│ ├── invoice-to-json PaLiGemma fine-tune, 88% extraction accuracy
│ ├── multimodal-ai Speech + text + image + video processing
│ ├── rlhf-dpo Direct Preference Optimization research
│ └── resume-parser OCR + ML multi-format resume parsing
├── cu-boulder/
│ ├── kl-divergence Statistical bridge with real-time viz
│ └── stellar-mapping CNN + ViT constellation detection
├── research/
│ ├── edge-aiot-inspection Microchip defect detection, 98% accuracy
│ └── autonomous-vehicle AGV with YOLOv4 on Jetson Nano
└── open-source/
├── roberta-toxicity Multi-label toxicity classifier
└── goodreads-prediction 1st place Kaggle, T5 generative model
4 directories, 12 projects
Zoho — 6 projects (click to expand)
| Project | What it does | Links | Stack |
|---|---|---|---|
| RAG Architecture | Advanced retrieval-augmented generation with hybrid search (5ms latency), multi-modal input, reduced hallucinations. Scaled to millions of users. | — | Python VLLM PyTorch FastAPI Redis PostgreSQL |
| Mamba-Transformer Hybrid | Novel architecture combining Mamba's selective state space models with Transformer attention for enhanced sequence modeling. Includes ablation studies. | — | Python PyTorch Mamba SSM JAX Wandb |
| Invoice Image to JSON | Fine-tuned PaLiGemma for structured data extraction from invoice images. Auto JSON schema validation and correction. | HF | Python PaLiGemma Transformers FastAPI |
| Multimodal AI System | Integrated speech, text, image & video processing. Meeting summarization, video scene understanding, cross-modal search & retrieval. | — | Python PyTorch Whisper CLIP TimeSformer |
| RLHF with DPO | Research exploring Direct Preference Optimization for RLHF. Comparative study with traditional approaches and advanced preference learning. | — | Python PyTorch Transformers Wandb Ray |
| Resume Parser | Multi-format resume parsing (PDF, DOC, images) with OCR, intelligent section recognition, date normalization, and entity extraction. | — | Python Tesseract spaCy TensorFlow Flask |
CU Boulder — 2 projects
| Project | What it does | Links | Stack |
|---|---|---|---|
| KL Divergence: A Statistical Bridge | Empirical investigation of KL divergence through binary classification with real-time visualization, VAE latent space exploration. | Code · Demo | Python PyTorch React Scikit-learn TensorFlow |
| Stellar Mapping | Deep learning system for automated constellation detection using CNN, Vision Transformer & EfficientNet ensemble across varying conditions. | Code · Demo | Python PyTorch CNNs ViT EfficientNet |
Research — 2 projects
| Project | What it does | Links | Stack |
|---|---|---|---|
| Edge AIoT for Product Inspection | CNN & SVM models for defective microchip identification with 98% accuracy. Cloud-based monitoring with automated reporting. | Demo | Python TensorFlow AWS DynamoDB Flask Raspberry Pi |
| Autonomous Ground Vehicle | AGV bot with real-time lane & object detection, autonomous navigation using Raspberry Pi, Jetson Nano, and YOLOv4. | Code · Demo | Python YOLOv4 ROS OpenCV Jetson Nano |
Open Source & Competitions — 2 projects
| Project | What it does | Links | Stack |
|---|---|---|---|
| RoBERTa Toxicity Classifier | Multi-label toxicity detection fine-tuned on RoBERTa for real-time content classification across multiple categories. | HF | Python PyTorch Transformers HuggingFace |
| Goodreads Rating Prediction | 1st place Kaggle solution using fine-tuned T5 generative model for book review rating prediction. State-of-the-art accuracy. | Code · Board | Python PyTorch T5 Pandas Scikit-learn |
____________________ sivakumar@boulder
| ________________ | -----------------
| | MS Data Sci | | OS: Life v24
| | CU Boulder | | Role: Applied AI Engineer @ Ricoh
| | GPA: 4.00 | | Prev: Data Scientist @ Zoho
| |________________| | Langs: Python, Java, TypeScript, R, SQL
| __ __ __ __ __ __ | ML: RAG, NLP, CV, Multimodal, RL
| |__|__|__|__|__|__|| Infra: AWS, FastAPI, PostgreSQL, Docker
| |__|__|__|__|__|__|| Models: 100M+ params trained
|____________________| Hobbies: Chess, Flute, Boxing
_|____________|_ Reads: Goodfellow, Jurafsky, Strang
/ ______________ \
/ / \ \
/_/ \_\
+---------------------------------------------+
| "Physics is my favourite, |
| Math is my queen, |
| Programming since 2018." |
+---------------------------------------------+