Full Stack Developer with 3 years of experience, building and shipping production-ready web applications end-to-end. Sharing my portfolio👇 🔗gitika-portfolio.vercel.app/ #FullStackDeveloper #BuildInPublic

The full architecture of Ledger — an AI invoicing agent. Auth → onboarding → 6-node LangGraph pipeline → invoice detail Invoice state is persisted: draft → confirmed → sent → paid Live at ledgerbrain.app

Built Ledger — an AI invoicing agent. Just type what you need. Under the hood: → LangGraph pipeline → Router → Generator → Editor → RAG → Multi-invoice support & Taxes back calculation → Vector search across past invoices → INR, USD, EUR More coming soon. #buildinpublic

Invoicing shouldn't take 30 minutes. Built something about that. The vision: describe your invoice in simple English — AI handles the rest. Demo dropping soon 👀 #buildinpublic #AI

I built a LangGraph agent with human approval today. Here’s how it works: 1️⃣ User enters a task 2️⃣ Agent generates a step-by-step plan 3️⃣ Workflow pauses for approval ✅ Approve → Agent executes the plan ❌ Reject → User is asked to run the agent again One thing I'm learning while building agents: Generation is easy. Control is the real challenge. LangGraph makes it much easier to design workflows where agents can pause, wait for humans, and then continue execution. #LangGraph #AIAgents #LangChain #BuildingInPublic

LangGraph becomes much easier once you understand these 3 concepts: 1️⃣ State Shared memory across the workflow 2️⃣ Nodes Actions like LLM calls or tools 3️⃣ Edges Logic that decides the next step Together they create stateful AI workflows with loops, retries, and branching. #AI #LangGraph #AIAgents #LearningInPublic

Many AI applications start with simple LLM chains. But real AI apps need more: - multiple steps - decision making - retries - loops - state across interactions LangChain simplifies building LLM apps with tools, memory, and structured pipelines using LCEL. But when workflows get complex, LangGraph helps build stateful AI agents with branching, loops, and decision-making. LCEL = pipelines LangGraph = agent workflows #AI #LangChain #LangGraph #AIAgents

After learning about RAG, Chunking, and Embeddings, I built a small Light RAG demo to see how it works in practice. Goal: Make the model answer questions only from the documents we provide instead of relying purely on its training data. How it works: Ingestion Phase - Add a document / policy text - Split it into chunks - Generate embeddings - Store them in a vector store Query Phase - User asks a question - Convert query → embedding - Retrieve similar chunks via semantic search - Send retrieved context to the LLM → generate answer Tech Stack: - Next.js - Node.js - LangChain - OpenAI Embeddings - In-memory Vector Store Building small demos like this is helping me understand how LLMs external knowledge power modern AI applications. #AI #RAG #LangChain #BuildInPublic #LearningInPublic

How do AI systems search through massive documents and find the right answer? Two key ideas make it possible: Chunking Large documents are split into smaller sections so the system can search specific pieces of information. Embeddings Each chunk is converted into a numerical vector that captures its meaning. When a user asks a question, the system compares the query embedding with stored chunk embeddings to find the most relevant context. This is the core mechanism behind RAG (Retrieval Augmented Generation) used in many modern AI applications. #AI #RAG #Chunking #Embeddings #BuildInPublic #LearningInPublic

Ever wondered how AI tools answer questions using your own documents? The answer is RAG AI retrieves relevant info first, then generates the answer. Docs → Chunking → Embeddings → Vector DB Query → Similarity Search → Context → LLM → Answer #AI #RAG #LLM #BuildInPublic

Built a Search Web Agent 🤖🔎 Next.js OpenAI LangChain Tavily. LLM answers simple queries. Automatically switches to web search when needed. Trying to explore how AI agents can combine LLMs real-time web data. #AI #LangChain #BuildInPublic

Building AI agents in production? LangChain simplifies: → Prompt management → Tool calling → Memory handling → Streaming responses → Multi-step reasoning With LCEL, complex workflows become clean, structured pipelines. #LangChain #LLM #AIEngineering #JavaScript

Built my first full-stack AI project: Promptify 🚀 React Node MongoDB OpenAI Chat image generation in one app. From learning concepts to building real products — this feels good. Learning in public. Building daily. #BuildInPublic #AI

Practiced OpenAI hands-on today 🚀 Built a basic AI chatbot from scratch to understand how LLMs actually work in an app. Small project, big clarity. #BuildInPublic #AI #OpenAI

Spent time today building clarity around modern AI fundamentals. LLMs, Generative AI, Agentic AI, AI agents, workflows, multi-agent systems, and RAG. Not deep dives — but enough to understand how intelligent systems are structured beyond simple prompting. #BuildInPublic #AI