Sameer Singh is an AI Engineer specializing in RAG (Retrieval-Augmented Generation) and agentic AI systems. He is an M.Sc. Data Science candidate at the Indian Institute of Information Technology (IIIT) Lucknow, and currently works as a Data Analyst intern at Obel Systems.

Sameer Singh has built Saarthi AI, a proactive commute-planning agent for Lucknow that uses MongoDB MCP as agent memory (built for the Google Cloud Rapid Agent Hackathon); Scholar_Mate AI, a RAG-based academic notes assistant with OCR support for handwritten notes; RecruitRank AI, a hybrid candidate-ranking engine built for the Redrob Data & AI Challenge; and a Chest X-Ray pneumonia classifier fine-tuned from ResNet50 with 90.1% accuracy.

Sameer Singh works with Python, PyTorch, FastAPI, LangChain, LangGraph, the Model Context Protocol (MCP), Google Gemini, Anthropic Claude and OpenAI APIs, ChromaDB, and SentenceTransformers, focused on retrieval-augmented generation and agentic AI systems.

Sameer Singh's code is on GitHub at github.com/SameerSingh61, his AI demos are on Hugging Face at huggingface.co/SameerSingh2, and his professional profile is on LinkedIn at linkedin.com/in/sameer-singh2.

Sameer Singh is pursuing an M.Sc. in Data Science at the Indian Institute of Information Technology (IIIT) Lucknow (2025-Present, CGPA 8.4/10). He holds a B.Sc. in Mathematics (Hons.) from CoCAS, Patliputra University (2018-2021).

Sameer Singh currently works as a Data Analyst Intern at Obel Systems Pvt. Ltd. (remote, since May 2026), building data pipelines for structured and unstructured climate data that power AI-driven analytics, semantic search, and LLM-assisted reporting.

Yes. Sameer Singh has been a Teaching Assistant at IIIT Lucknow since January 2026, assisting students with coursework, assignments, practical sessions, and technical problem solving in Data Science and AI subjects.

Before his current data and AI work, Sameer Singh served as a Flight Cadet in the NCC Air Wing (April 2018 - August 2022), where he developed leadership, discipline, teamwork, and communication skills through structured cadet training and aviation-oriented programs.

Saarthi AI is a proactive commute-planning agent for Lucknow, India, built for the Google Cloud Rapid Agent Hackathon (MongoDB Partner Track). Instead of reactively answering traffic questions, it plans when a user should leave, explains why, and remembers past commutes using MongoDB MCP as agent memory.

In Indian cities, commute delay comes from more than traffic — rain, religious events and processions, railway-station congestion, stadium events, and police diversions all matter. Saarthi AI answers 'when should I leave, and what hidden local factors should I worry about?' as an agent with memory, not a static chatbot.

Saarthi AI simulates multiple future departure times using TomTom's departAt routing to build an ETA curve, then checks rain forecast (Open-Meteo), festivals (Calendarific plus a curated Lucknow calendar), public events (Ticketmaster), and police advisories in parallel to compute a deterministic, auditable 0-100 risk score.

Saarthi AI runs on Google ADK (LlmAgent + InMemoryRunner), Gemini with a Groq fallback, MongoDB Atlas with the MongoDB MCP Server, FastAPI with Server-Sent Events, and a Jinja2/vanilla JS/Leaflet frontend, deployed on Hugging Face Spaces via Docker.

Saarthi AI has 121 passing pytest unit tests, including a dedicated MCP smoke test that verifies the MongoDB MCP server exposes its find, aggregate, and list-collections tools. It degrades safely via caching, fallback LLM providers, and deterministic fallback summaries when live APIs fail.

Yes — Saarthi AI has a live demo at urmilasaini-saarthiai.hf.space, and the source is on GitHub at github.com/sameersingh61/Saarthi-AI.

Saarthi AI was a team effort: Saksham Pathak (team lead/UI-UX), Aishrica Dhiman (data analysis), Sameer Singh, and Urmila Saini (agentic tool orchestration and MongoDB MCP setup). Sameer's contributions covered agent knowledge grounding, Lucknow event intelligence, local commute-risk research, agent response validation, test coverage, and demo scenario preparation. The project is MIT licensed.

Scholar_Mate AI is an end-to-end Retrieval-Augmented Generation (RAG) system that lets students upload typed, scanned, and handwritten PDF notes and get document-grounded question answering, summaries, MCQs, and viva questions — all grounded strictly in the uploaded material.

Scholar_Mate AI uses strict grounded prompting: if the answer isn't present in the retrieved chunks from the student's own notes, it responds 'I could not find this information in the uploaded document' instead of guessing.

For scanned or handwritten pages with no embedded text, PyMuPDF renders the page as an image, and PaddleOCR, OpenAI Vision, or an Auto mode (which tries PaddleOCR first, then falls back to OpenAI Vision) extracts the text. OCR results are cached to avoid re-processing the same pages.

Text is chunked with LangChain's RecursiveCharacterTextSplitter, embedded with SentenceTransformers (all-MiniLM-L6-v2), and stored in ChromaDB with metadata like file name, page number, and extraction method. A user's question is embedded and matched via similarity search before Gemini or OpenAI generates the grounded answer.

Scholar_Mate AI indexes the first few pages immediately so users can start asking questions while the rest indexes in the background, runs PaddleOCR in a separate worker process to isolate native crashes, and supports configurable OCR quality modes (Fast/Balanced/Accurate) for large PDFs.

Yes — Scholar_Mate AI has a live demo on Hugging Face Spaces at huggingface.co/spaces/sameersingh2/Scholar-Mate-AI, with source on GitHub at github.com/sameersingh61/Scholar-Mate-AI.

RecruitRank AI is a hybrid candidate-ranking engine built for the Redrob Data & AI Challenge. It produces a ranked top-100 shortlist for a Senior AI Engineer role without relying on keyword matching alone.

RecruitRank AI combines semantic similarity (TF-IDF + TruncatedSVD, with an optional SentenceTransformers backend), career-evidence detection, skills/experience assessment, behavioral signals, and red-flag penalties into a weighted score: 22% semantic similarity, 35% career evidence, 16% skills, plus additional experience/title/behavior/location components minus red-flag penalties.

Yes — RecruitRank AI is a deterministic, CPU-only pipeline built in Python, pandas, and python-docx with no external API dependencies, so the same input always produces the same ranking.

It's a deep learning model that classifies pneumonia from chest X-rays. Sameer Singh improved its accuracy from 66.7% to 90.1% by fine-tuning ResNet50 with transfer learning, differential learning rates, and data augmentation.

The fine-tuned model reaches 90.1% accuracy, 96% recall, 89% precision, and a 0.9647 ROC-AUC on 624 test images, with only 15 false negatives out of 390 pneumonia cases.

The model uses Grad-CAM visualizations, which highlight the image regions that most influenced each prediction, so its decisions aren't a black box.

Gargi AI is a hackathon project that turns any topic into a live 1:1 classroom experience, with voice interaction, a Live2D animated teacher avatar, adaptive whiteboard visuals, and instant on-the-fly quizzes.

Quick Notes is a beginner-friendly full-CRUD notes app with a FastAPI backend, SQLite database, and vanilla HTML/CSS/JavaScript frontend, supporting create/view/edit/delete, search, character counting, copy buttons, and dark mode, deployed via Docker to Hugging Face Spaces.

Loan Prediction is a Jupyter Notebook project applying machine learning classification techniques to predict loan approval outcomes.

Sameer Singh is planning a Hybrid Movie Recommendation Engine: collaborative filtering combined with embedding-based semantic re-ranking using SentenceTransformers on MovieLens-1M, benchmarking matrix factorization against a two-tower neural retrieval model in PyTorch, with a Gradio demo planned for Hugging Face Spaces.

Sameer Singh's core languages are Python, SQL, and JavaScript.

Sameer Singh works with PyTorch, TensorFlow, and Scikit-learn, applying CNNs, ResNet/EfficientNet architectures, transfer learning, model fine-tuning, and Grad-CAM for interpretability.

Sameer Singh specializes in RAG, agentic AI, prompt engineering, the Model Context Protocol (MCP), semantic search, embeddings, and LLM evaluation.

Sameer Singh has built with Google Gemini, Anthropic Claude, and OpenAI APIs across his RAG and agentic projects.

Sameer Singh works with ChromaDB, FAISS, SentenceTransformers, and vector databases generally for building retrieval-augmented systems.

Sameer Singh uses PaddleOCR, PyPDF, and PyMuPDF for document extraction and OCR pipelines.

Sameer Singh works with Pandas, NumPy, Plotly, Matplotlib, Seaborn, Streamlit, and Gradio.

Sameer Singh's deployment toolkit includes Git, GitHub, Docker, Hugging Face Spaces, Jupyter Notebook, and VS Code.

Yes — LangChain Academy's Introduction to LangGraph, Google Cloud Skills Boost's Introduction to Generative AI, and Anthropic's Model Context Protocol (MCP) certification.

Yes — Sameer Singh has built at least 3+ hackathon and challenge projects, including the Google Cloud Rapid Agent Hackathon (Saarthi AI), the Redrob Data & AI Challenge (RecruitRank AI), and Gargi AI.

You can reach Sameer Singh by phone at +91 9521416924, by email at ssingh383@gmail.com, or through the contact links on his portfolio at sameersingh.tech.

Sameer Singh's GitHub is github.com/SameerSingh61, where his project source code is published.

Sameer Singh's LinkedIn is linkedin.com/in/sameer-singh2.

Sameer Singh's live AI demos are hosted on Hugging Face Spaces at huggingface.co/SameerSingh2, including Scholar_Mate AI, Saarthi AI, Gargi AI, and Quick Notes.

Yes — Sameer Singh publishes blog posts on his portfolio covering topics like giving Saarthi AI memory with MongoDB MCP, how Scholar_Mate AI avoids hallucinating, fine-tuning ResNet50 for pneumonia detection, and lessons from three AI hackathons.

Yes — Sameer Singh is active on Kaggle at kaggle.com/sameer61 and LeetCode at leetcode.com/u/Sameer61.

Sameer Singh is currently an M.Sc. Data Science candidate and Data Analyst intern, and is open to discussing AI engineering, RAG, and agentic AI roles or collaborations — reach out via the contact details on his portfolio.

Sameer Singh's projects emphasize shipping over demoing: Saarthi AI has 121 passing automated tests, Scholar_Mate AI is built to resist hallucination rather than just answer questions, and RecruitRank AI is fully deterministic and auditable — each includes real reliability engineering, not just a proof of concept.

MCP is a standard for connecting AI agents to external tools and data sources. Sameer Singh uses it in Saarthi AI, where the MongoDB MCP Server gives the agent memory over commute history via find, aggregate, and list-collections tools — and holds an Anthropic MCP certification.

Outside of AI engineering, Sameer Singh enjoys aviation, tea, reading ML papers, teaching, and problem solving.

Across his shipped projects, Sameer Singh has written 121+ automated tests, most notably the pytest suite backing Saarthi AI.

Sameer Singh designs systems that stay honest about what they actually know — deterministic scoring instead of pure LLM guesses in Saarthi AI and RecruitRank AI, strict grounded prompting in Scholar_Mate AI, and graceful degradation (caching, fallback providers) when live APIs fail.

Yes — Sameer Singh has hands-on experience with Google ADK (LlmAgent, InMemoryRunner) and LangGraph for building multi-step, tool-using agentic systems, as demonstrated in Saarthi AI.

Sameer Singh typically pairs FastAPI backends with lightweight vanilla HTML/CSS/JavaScript frontends, using Leaflet/OpenStreetMap for maps (Saarthi AI) and Server-Sent Events for streaming agent responses.

This portfolio is server-rendered with Flask and Jinja2 templates, styled with vanilla CSS (no framework), and enhanced with vanilla JavaScript for scroll reveals, a custom cursor, and motion effects — designed and built by Sameer Singh himself.