Hi, I'm Samarth Singh
Building production-ready GenAI systems with RAG and LLMOps. Specialized in semantic search, fine-tuning transformers, and deploying intelligent LLM applications with retrieval architectures.
About Me
Samarth Singh
GenAI & RAG Engineer
GenAI Engineer & Research Enthusiast
BTech Computer Science student at VIT Bhopal (CGPA 8.45) specializing in Generative AI, RAG systems, and LLMOps. Proven expertise in building production-ready LLM applications with semantic search, hybrid retrieval (BM25 + dense vectors), and parameter-efficient fine-tuning (LoRA/PEFT). Experienced in deploying transformer models with evaluation frameworks achieving 89%+ correctness.
Passionate about making LLMs more accurate and grounded through intelligent retrieval architectures. Published fine-tuned models on Hugging Face with comprehensive evaluation (ROUGE, BERTScore, METEOR). Seeking GenAI/LLMOps internships to build scalable AI systems that solve real-world problems with measurable impact.
Years Experience
Projects Completed
CGPA
The tools, languages, and frameworks I leverage to build powerful and innovative web applications.
LLMOps & GenAI
LangChain
Proficiency Level
LangGraph
Proficiency Level
Agno
Proficiency Level
Vector Databases
Proficiency Level
FastAPI
Proficiency Level
Groq
Proficiency Level
Hugging Face
Proficiency Level
Prompt Engineering
Proficiency Level
MLflow & W&B
Proficiency Level
Gradio/Streamlit
Proficiency Level
ML/DL & AI Engineering
PyTorch
Proficiency Level
TensorFlow
Proficiency Level
Scikit-learn
Proficiency Level
Transformers
Proficiency Level
Pandas/NumPy
Proficiency Level
Jupyter
Proficiency Level
Programming Languages
Python
Proficiency Level
C++
Proficiency Level
JavaScript
Proficiency Level
TypeScript
Proficiency Level
SQL
Proficiency Level
Full-Stack Development
Next.js
Proficiency Level
Node.js
Proficiency Level
PostgreSQL
Proficiency Level
React
Proficiency Level
Tailwind CSS
Proficiency Level
Tools & DevOps
Git/GitHub
Proficiency Level
Docker
Proficiency Level
AWS
Proficiency Level
Linux/CUDA
Proficiency Level
Education & Certifications
My academic journey and professional development
CGPA: 8.45/10 | Focus on Artificial Intelligence, Machine Learning, and Cloud Computing
Key Courses
Coding Activity
My development metrics and problem-solving statistics
Featured Projects
Explore my latest work showcasing my skills and expertise
Argus — Autonomous Research Engine
Production-grade multi-agent research pipeline using LangGraph supervisor pattern with 5 specialist agents (planner, researcher, critic, writer, supervisor) orchestrated via LLM-driven routing. Async job architecture with SQLite persistence and LangGraph checkpointing — research jobs survive agent failures and every LLM call is traced end-to-end in LangSmith. Integrates Tavily, ArXiv, and Wikipedia to synthesize cited markdown reports in 30–90 seconds.
DoCopilot - RAG Document Q&A System
Production-grade RAG application with hybrid search (BM25 + dense vectors) using Qdrant and reranking. Achieved 89.2% correctness, 90.5% relevance, 100% source grounding on 40-query evaluation with guardrails for PII redaction and prompt injection detection. Built full-stack with Next.js frontend and FastAPI backend, processing PDFs/TXT with 2.86s average latency.
FLAN-T5 Dialogue Summarizer
Fine-tuned FLAN-T5-base with LoRA on SAMSum dataset (14.7K dialogues), achieving 49.01 ROUGE-1, 72.25 BERTScore F1, and 42.51 METEOR scores. Implemented parameter-efficient training updating only 2% of parameters with FP16 mixed precision. Deployed interactive Gradio app on Hugging Face Spaces with configurable beam search and published model with reproducible evaluation.
RoBERTa Banking Intent Classifier
Fine-tuned RoBERTa-base on Banking77 dataset (77 intents, 13K queries) achieving 93.7% accuracy and 93.6% macro-F1. Implemented standard transformer fine-tuning with AdamW optimizer, weight decay, and FP16 training on GPU. Added experiment hygiene with fixed seeds, consistent tokenization, epoch-level metrics tracking, and best-checkpoint selection for robust evaluation.
Project Loom
Full-stack project-sharing platform with Next.js leveraging SSR and ISR, reducing page load times by 50%. Architected scalable backend with Sanity.io headless CMS managing 1,000+ project entries. Implemented secure authentication with NextAuth.js and PostgreSQL, enabling users to manage profiles, post projects, and interact with content.
Modern Portfolio
Personal portfolio website built with Next.js, TypeScript, and Tailwind CSS featuring modern UI elements, smooth animations with Framer Motion, and optimized SEO for GenAI/RAG internships. Implements dark mode, responsive design, and accessibility best practices with Lighthouse scores 90+ across all metrics.
Dexplorer
Interactive Pokémon discovery web application for searching and exploring the original 150 Pokémon with detailed information, stats, and type filtering. Built with React and modern JavaScript, featuring responsive design and smooth user experience.
Blog & Writing
Deep-dives into production AI systems, architecture decisions, and lessons learned the hard way.
DoCopilot: Building a Production-Grade RAG System with Hybrid Search, Reranking, and Safety Guardrails
How I went from a basic chatbot to an 89%+ accurate document QA system — with ablation studies to prove it. Most RAG tutorials stop at 'embed your PDF, do cosine similarity, feed to GPT.' That's fine for a demo. It breaks in production.
The Power of Normalization: How Feature Scaling Transforms Neural Network Performance on Tabular Data
A simple standardization step can transform a completely broken model (R² ≈ 0) into a high-performing one (R² > 0.8) — while dramatically reducing training time. The 160,800x improvement in R² score proves normalization is not optional.
Achievements & Certifications
Awards, certifications, and milestones in my journey
Completed specialization in supervised/unsupervised learning, feature engineering, model evaluation, and scikit-learn for practical ML applications.
2025
FLAN-T5 Summarizer with reproducible evaluation achieving 49.01 ROUGE-1 and 72.25 BERTScore F1 on SAMSum dataset
Oct 2025
Maintained 8.45 CGPA with focus on AI/ML coursework including DSA, Cloud Computing, and Software Engineering
2023-Present
My Approach
Planning & Strategy
We'll collaborate to map out your website's goals, target audience, and key functionalities. We'll discuss site structure, navigation, and content requirements.
Development & Progress
Once we agree on the plan, I cue my lofi playlist and dive into coding. From initial sketches to polished code, I keep you updated every step of the way.
Deployment & Launch
After thorough testing and your final sign-off, I deploy your project to production with CI/CD pipelines, monitoring, and a smooth launch. Post-launch support ensures everything runs flawlessly from day one.
Future Enhancements
Planned features and improvements — turning this portfolio into a continuously evolving product.
Agentic Portfolio Assistant
A LangGraph-powered chat agent embedded in the portfolio — visitors can ask questions about projects, get technical details, or request a tailored summary. RAG over project READMEs + resume as the knowledge base.
Live Project Dashboards
Real-time evaluation dashboards for deployed models — live ROUGE/BERTScore trends for the FLAN-T5 summarizer, accuracy drift detection for the RoBERTa classifier, and latency telemetry for Argus.
Multi-Language Support (i18n)
Internationalise the portfolio with next-intl — Hindi, Spanish, Japanese. Auto-detect browser locale and serve locale-specific OG images and structured data for broader reach.
Interactive ML Playground
Embed live Hugging Face Spaces iframes directly into project cards — let visitors run the FLAN-T5 summarizer or RoBERTa classifier without leaving the portfolio.
Auth & Guestbook
Add NextAuth.js GitHub OAuth so visitors can leave verified notes in a public guestbook. Messages stored in Postgres via Prisma — a small but memorable personal touch.
CMS-Backed Projects & Blog
Move projects and blog posts to a headless CMS (Sanity.io or Contentlayer) so new content can be published without code deployments. MDX support for rich blog articles.
Open to collaboration — reach out if any of these interest you
Contact Me
Have a project in mind or want to collaborate? Feel free to reach out!