// ML Engineer · GenAI Developer · AI Engineer

Your product ships features.
Mine ships intelligence.

I build end-to-end ML systems and LLM-powered applications — RAG pipelines, agentic workflows, reasoning interfaces. From LangGraph orchestration to production FastAPI deployments, I close the gap between research and real users.

View Projects → Get in Touch
Scroll to explore

// The Problem I Solve

Every AI product hits the same three walls.

WALL 01
INTELLIGENCE

LLMs hallucinate. RAG fixes that — if you build it right.

Generic LLM calls break in production. I engineer retrieval pipelines with FAISS, ChromaDB, and proper chunking strategies so answers are strictly grounded in your data.

WALL 02
RELIABILITY

Agents that work in notebooks fail in the real world.

I build LangGraph-orchestrated agents with state machines, intent classifiers, and fallback paths — designed to handle edge cases before your users find them first.

WALL 03
DEPLOYMENT

Great models stuck behind bad infra ship nothing.

FastAPI backends, Docker containers, AWS EC2 deployments, Redis caching — I take ML from Jupyter notebooks to production-grade APIs that actually scale.

// Who I Am

The full-stack AI engineer your team is missing.

I'm Adarsh Pandey and I have deep hands-on expertise across the entire AI stack — from neural network fundamentals to deploying multi-agent systems in production. Certified by Anthropic, Google, and Kaggle. Obsessed with building things that actually work.

01 — GenAI & LLMs

Agent Intelligence. Every tool. Every chain. Orchestrated.

LangChain, LangGraph, LangSmith — the full agentic stack. RAG pipelines with FAISS & ChromaDB. Groq API, Gemini API, Ollama for local deployment. Prompt engineering that gets results.

// Generative AI Stack

Orchestration

LangChain LCEL LangGraph LangSmith

Vector Stores

FAISS ChromaDB HuggingFace Embeddings

LLM APIs

Groq API Gemini API Ollama

Patterns

RAG Pipelines Prompt Engineering Agents & Memory

02 — ML & Deep Learning

From gradient descent to production models.

Supervised & unsupervised learning, XGBoost, Random Forest, CNNs, RNNs, LSTMs, Transformers, HuggingFace. Feature engineering, cross-validation, evaluation metrics — the full pipeline.

// ML & Deep Learning Stack

Frameworks

PyTorch TensorFlow HuggingFace Scikit-learn

Models

XGBoost Random Forest LSTM / GRU Transformers

Data

NumPy Pandas Matplotlib Power BI

03 — Backend & DevOps

Ship it. Build it. Scale it.

FastAPI, Node.js, Express.js, REST APIs, WebSockets, MongoDB, MySQL, Redis, Docker, AWS EC2, Nginx. Full production-grade backend engineering.

// Backend & Infrastructure

APIs & Servers

FastAPI Node.js Express.js Streamlit

Databases

MongoDB MySQL Redis SQLite

DevOps

Docker AWS EC2 Nginx Git / GitHub

// Projects

Production-grade AI.
Not just notebooks.

01

Axiom — Multi-Tool Agentic Chatbot

LangGraph-orchestrated AI agent with per-thread PDF RAG (FAISS + all-MiniLM-L6-v2), DuckDuckGo web search, Alpha Vantage stock lookup, and calculator — dynamically routed by Llama 3.3 70B via Groq. Persistent multi-thread memory via SqliteSaver, auto-generated chat titles, and full conversation restore.

LangGraphFAISSHuggingFaceGroqStreamlitLangSmithSQLite
GitHub ↗
02

AutoStream AI Sales Agent — Agentic LLM Workflow

Production-style LangGraph sales agent with a two-node directed graph (detect_intent → generate_response) managing persistent TypedDict state across 5+ conversation turns. LLM-powered 3-class intent classifier decoupled from response generation, with state-machine-enforced lead capture that prevents premature execution by design.

LangGraphLangChainGroq APILlama 3.3 70BStreamlitRAG
GitHub ↗
03

YouTube RAG Chatbot — LLM-Powered Video Q&A

Full RAG pipeline using LangChain LCEL for semantic Q&A over any YouTube video's transcript. FAISS vector store with all-MiniLM-L6-v2 embeddings (chunk_size=1000, overlap=200), top-k retrieval, multilingual transcript handling, and conversational memory via last-3-exchange injection into every prompt.

LangChain LCELFAISSGroq APIYouTube Transcript APIStreamlit
GitHub ↗
04

Waffle Chat — Conversational AI with Reasoning Transparency

Multi-turn conversational AI interface powered by DeepSeek-R1-0528 reasoning LLM via LangChain and HuggingFace Inference API. Regex-based <think> block parsing to separate model reasoning from final answers, with a toggleable reasoning panel. Persistent multi-turn memory and optimized loading via @st.cache_resource.

DeepSeek-R1LangChainHuggingFace APIStreamlitGroq
GitHub ↗

// Early Numbers

4 production-grade agents.
Here's what I built.

5+

end-to-end LLM-powered applications built and deployed

10+

AI/ML frameworks mastered across the full stack

5

industry certifications from Anthropic, Google, Kaggle, and HP

// Experience

Production exposure.
Not just coursework.

CloudErrand.in

Feb 2026 –
April 2026

Cloud Integration
Intern

Cloud Integration Intern

  • Engineered 10+ production-grade integration workflows on Celigo iPaaS, connecting enterprise systems including NetSuite ERP with external platforms — automating end-to-end data pipelines that eliminated manual data entry across live client environments.
  • Contributed to 5+ real client implementations, handling full project lifecycle from requirement analysis to production deployment — gaining direct exposure to enterprise delivery standards, timelines, and troubleshooting in live systems.
  • Designed and deployed a multi-branch automation flow capable of intelligently routing and processing 5+ NetSuite record types (Customer, Sales Order, Purchase Order, and more) from a single JSON input — reducing record creation effort from a manual multi-step process to a fully automated, zero-touch workflow.
  • Architected API-based integrations using NetSuite's REST/SOAP APIs and Celigo connectors — implementing authentication, field mapping, data transformation, and error-handling mechanisms to ensure reliable, real-time data synchronization across cloud platforms.
  • Standardized workflow automation best practices across integration projects — including structured naming conventions, modular flow design, audit logging via custom NetSuite Record Holder tracking records, and monitoring — contributing to maintainable and scalable integration architecture.

// How I Work

Discover. Build. Ship.
Repeat.

1
Discover
2
Build
3
Ship
Cursor · Adarsh's Dev Environment
$ git clone adarsh/new-agent && cd new-agent

Architecture Discovered

RAG pipeline + 3 LangGraph nodes + 2 tools
1 unguarded reasoning path detected
Adding intent classifier guard node
$ docker build -t agent:prod && docker push

Deployment Status

FastAPI backend live on AWS EC2
FAISS index built · 10K vectors indexed
Redis cache warm · avg latency 180ms
LangSmith tracing active
$ echo "shipped." ✓ shipped.

// Certifications

Certified by the companies
building the future.

Anthropic

Claude Code & Model Context Protocol (MCP)

Issued March 2026 · AI agent workflows, MCP architecture, AI-assisted coding

Kaggle × Google

5-Day AI Agents Intensive Course

Issued December 2025 · AI Agents, Generative AI, LLMs, Prompt Engineering

Google for Education

Gemini Certified University Student

Issued November 2025 · Valid November 2028 · Google Workspace, Data-Driven Collaboration

HP LIFE (HP Foundation)

AI & Data Science Certifications (4×)

Issued 2025 · AI for Beginners, Data Science & Analytics, AI for Business, Agile

// Achievements

Built in competitions
Validated by industry.

2× Unbeatable Coding Champion — Makhanlal Chaturvedi National University

2x champion of coding competition consecutively, competed against multiple competitors and we all were restricted to basic technologies for implementations. Restricted to plan, acquire perfect resources from given pool of resources and implementation of our thoughts and then testing whether it works as expected and then to present infront of expert within only 90 minutes.

Kaggle × Google

5-Day AI Agents Intensive Course

Completed all 5 days of the intensive program co-hosted by Kaggle and Google, earning the official badge. Covered agentic AI architectures, tool use, and multi-agent systems — directly aligned with my ongoing LangGraph and MCP work.

PPT Design Competition Winner — Makhanlal Chaturvedi National University

Secured 1st position in the university-level presentation competition, demonstrating skills in structured storytelling, visual communication, and technical clarity.

Let's build something that actually works.

Open to ML Engineer, GenAI Developer, and full-stack AI roles.
Also open to interesting freelance projects and collaborations.