Headline

Applied AI Engineer | Forward Deployed Engineer | Python | Building production AI agents for customer-facing businesses | Agentic AI | RAG | LangChain | LangGraph | Vector Databases | Voice AI | Pipecat | AWS

About

I've always been an engineer at heart — I just spent eight years running a business first. And that turned out to be the best engineering education I could have asked for, because it buried me in real problems with real consequences. I ran a multi-brand retail business and scaled it from 2 stores to 5. Along the way I wore every hat — sales, marketing, finance, accounts, operations, and managing a team of 15–20 people. I kept running into the same thing: messy data I had to turn into decisions, workflows that broke as we grew, tools that didn't fit how the work actually happened, and people and customer challenges that no software solved. The difference was that I looked at these problems differently than most business owners — because I'm a genuine tech enthusiast who's never been able to use something without wanting to know how it works under the hood. I instinctively thought like an engineer: I wanted to understand why something broke and how to fix it at the root, not just patch it. So I started building. Since 2022, alongside the business, I taught myself the stack — deployed and kept our production server live on AWS EC2 for 4+ years, automated our operations across HubSpot, Zapier, and Make.com, and shipped our website on React/TypeScript. Then I went deep on AI: machine learning, deep learning, and NLP, and got hands-on with RAG, vector databases (FAISS, ChromaDB, Pinecone), LangChain, LangGraph, AutoGen, and real-time voice agents on Pipecat. Today I build AI agents for the kind of messy, customer-facing problems I used to live. What I bring that most engineers don't: I've already been the customer. I've felt the frustration of a tool my team wouldn't adopt, sat across from unhappy clients and earned their trust back, and owned the outcome on a P&L. Running sales for years also taught me how to genuinely understand what a customer needs and build it for them — which, it turns out, is most of the job in Forward Deployed and Applied AI roles. I'm still very much learning, endlessly curious about how things work, and building in the open — enjoying every bit of it. What I'm building: Voice AI sales agent — real-time intent detection, NL-to-SQL over live product data, multi-turn memory Agentic data analyzer — a planner–executor multi-agent system (AutoGen) that writes and runs its own code in a sandbox Open to Applied / Agentic AI Engineer, Forward Deployed Engineer, and Conversational AI roles — India-based or remote.

Location

Noida

AI stack

antigravity

claude code

gemini

grok

hugging face

kimi

langchain

langgraph

langsmith

livekit

mcp

n8n

notebooklm

obsidian

openai

perplexity

claudecode

huggingface

AutoGen

ChatGPT

Chroma

Gamma

Google ADK

HeyGen

Kimi Code

Langflow

LlamaIndex

Microsoft Copilot

OpenAI Agents SDK

OpenAI Playground

Pipecat

Pinecone

Ragas

Supabase

Tavily

Vercel AI SDK

Zapier

Work experience

Applied AI Engineer (Freelance) at Self-Employed, Jan 2026 - Present · 7 mos, Building production-grade Generative AI systems — RAG pipelines, multi-agent architectures, and LLM-powered business automation. Self-directed, full-time build. ● RAG Pipeline — End-to-end retrieval-augmented generation (LangChain + vector search over FAISS / ChromaDB) with embeddings, optimised chunking, and query expansion; tuned to sub-2s responses across 500+ test queries on a custom evaluation harness. ● Multi-Agent System — Planner–Executor architecture (AutoGen): an LLM generates Python from natural language, executes it in a sandbox on live CSV/Excel data, and synthesises results — ~60% less manual analysis effort in testing. ● Voice AI Sales Agent — Agentic assistant (LangChain + LangGraph) with real-time intent detection, dynamic tool-calling, NL-to-SQL over a live product database, and context-aware multi-turn conversations. ● Conversational Memory — Persistent multi-turn memory with a modular pipeline designed for real business deployment. ● Currently building — Production RAG over real retail documents with vector search, evaluation (faithfulness / retrieval metrics), and guardrails / fallbacks for grounded, cited answers. Stack: Python · LangChain · LangGraph · AutoGen · RAG · Vector Databases (FAISS, ChromaDB, Pinecone) · Embeddings · NL-to-SQL · FastAPI · OpenAI API · Hugging Face · Streamlit · scikit-learn · AWS

Managing Director at Skope Infonet Pvt Ltd, Apr 2016 - Dec 2025 · 9 yrs 9 mos, Noida, Dell, Lenovo & Asus exclusive retail stores | Scaled from 2 to 5 stores | Multi-crore annual revenue | Team of 15–20. I ran the business end-to-end and also served as the company's sole technology builder. Over the last few years (from 2022) I went deeper — architecting, deploying, and maintaining the company's systems myself using modern and AI-assisted tools. ● AWS Cloud Infrastructure — Independently architected and deployed the company's production server on AWS EC2 (security groups, S3, networking); live and operational for 4+ years. ● React / TypeScript Website — Built and deployed skopeinfonet.com (still live) in React + TypeScript using modern AI-assisted tooling (Bolt.new), hosted on Netlify. ● E-Commerce Platform — Built the company store with product catalogue, cart, and WooCommerce checkout; handled hosting, payment-gateway integration, and maintenance end-to-end. ● CRM & Workflow Automation — Designed a HubSpot + Mailchimp stack integrated via Zapier and Make.com; automated lead capture, email campaigns, and lifecycle triggers across CRM, inventory, and reporting — ~40% less manual follow-up. (Directly analogous to modern LLM tool-chaining and agent workflows.) ● Data & Decisions — Built Excel / Google Sheets dashboards with automated logic for sales, inventory, and financial reconciliation across all stores; automated lead categorisation and analytics reduced missed follow-ups ~30%. ● Business Outcomes — Scaled 2→5 stores at multi-crore revenue with 95% employee retention; owned direct relationships with Dell, Lenovo, and Asus and daily client interaction across sales, finance, and operations.

Associate Software Engineer at Cognizant Technology Solutions, Aug 2015 - Aug 2016 · 1 yr 1 mo, Bengaluru, Supported 24×7 network and system monitoring operations for LyondellBasell (LYB), a global chemical manufacturing enterprise. ● Monitored enterprise network infrastructure using internal and third-party tools to ensure continuous uptime and performance stability. ● Gained early exposure to large-scale enterprise IT operations and cross-functional technical support workflows.

Education

Advanced Certification in Software Engineering for Cloud, Blockchain & IoT, CODE IIT Madras

Post Graduate Diploma in Business Management, Marketing/Marketing Management, General, Narsee Monjee Institute of Management Studies, Mumbai

Public links

github: https://github.com/Shivai20