NR
NR* — 001 · Jersey City, NJOpen to full-time roles

● Issue 01 · an introduction

Hi — I’m Neeraj.

I build AI & Full-Stack systems end-to-end — from Agentic MCPs and retrieval to forward-deployed solutions and the interfaces they live inside.

Production RAG over 10K+ docs, ReAct agents on GPT-4 function-calling, RAGAS + LangSmith eval pipelines, and the streaming UIs that ship them. Currently in Jersey City, NJ — open to AI, Agentic, Forward-Deployed, or Full-Stack roles.

See the workGet in touch
Scroll · 02 · About

Issue 02

One engineer, the entire AI stack.

Building products from the data pipeline up to the user interface, without dropping context across teams.

“Most teams hand a model to one person, the API to another, and the UI to a third. I’m the full vertical — retrieval, agents, evaluation, the streaming interface, the AWS pipeline that ships it. Fewer tickets. Faster iteration. Production behaviour you can actually measure.”

Currently shipping production RAG and agentic systems at T-Mobile — hybrid retrieval over 10K+ docs, ReAct agents on GPT-4 function-calling, and a 50-template prompt library with measured hallucination cuts of ~60%.

Before that: 2 years at Wipro on enterprise NLP, an MS in CS at NJIT (2025), and a steady stream of open-source AI tools — VIBETTER (MCP for IDEs), Context Refinery (LangGraph prompt optimizer), and the rest of the case studies below.

Four principles

i.

Ship systems, not demos.

Every project is deployed, monitored, logged, and behind CI/CD — or it doesn't count.

ii.

Evaluation isn't optional.

If retrieval quality and hallucination rate aren't measured, you're guessing in production.

iii.

Accessibility is a feature.

WCAG compliance, keyboard paths, and screen-reader support get shipped with the MVP.

iv.

Infrastructure is product.

Deployment, observability, and reliability are part of the user experience — not afterwards.

Outside the terminalPhotographyGymTravelAnimeMotorcyclesCarsBuilding things end-to-end

Issue 04

AI products, shipped end-to-end.

From proof-of-concept to production — robust pipelines, evaluated retrieval, and fluid interfaces.

01

CRE Intelligence Platform

NYC Agentic Underwriter & Feasibility

2026
02

VIBETTER

Open-Source MCP Server for AI Coding IDEs

Feb 2026
03

Context Refinery

LangGraph Multi-Agent Context Orchestration Engine

2026
04

Production RAG System

Enterprise Knowledge Retrieval & LLM Evaluation at T-Mobile

Jan 2025
05

Autonomous AI Agent

Multi-Step Reasoning with Tool Use & Cited Reports

2025
06

AI Health Platform

RAG Chatbot & Gamified Nudges — NJIT Capstone

2025
07

AI Expense Management

NLP Categorization & Anomaly Detection

2025
08

ML Image Classification

ResNet-50 Fine-Tuning & Interpretability

2024
All work index

Issue 03

A working history.

Three roles, one pattern — take an ambiguous AI problem, ship a measurable system, instrument it, and keep shipping.

Jan 2025 – Present

Remote · USA

AI Software Engineer — RAG, Agents & Eval — T-Mobile (via Wipro)

  • Shipped a production RAG pipeline over 10K+ enterprise documents — LangChain + ChromaDB + FastAPI, sub-2s p95 retrieval, hybrid BM25 + dense with cross-encoder re-ranking
  • Built an autonomous ReAct agent on GPT-4 function-calling — 5–10 tool calls per query across CRM, billing, and knowledge APIs, deflecting Tier-1 support load
  • Stood up an LLM evaluation framework with RAGAS + LangSmith — measured faithfulness, context recall, and hallucination rate; cut hallucinations ~60% across the assistant
  • Authored 50+ prompt templates with safety guardrails, tone alignment, escalation logic, and PII redaction — reviewed and version-controlled like code
  • Built the streaming chat UI in Vue 3 + React + Tailwind — token-level rendering, retry/resume, full WCAG 2.1 AA keyboard and screen-reader support
  • Deployed services as Dockerized FastAPI microservices on AWS — zero-downtime rollouts via GitHub Actions, structured logs to CloudWatch, health/readiness probes

Aug 2023 – Present

Remote

AI Engineer — Open Source & Contract — Independent

  • Built and open-sourced VIBETTER — an MCP server that gives AI coding IDEs (Claude Code, Cursor) richer project context via stateful tool use
  • Built Context Refinery — a LangGraph multi-agent prompt optimizer with a Tauri desktop UI; iteratively rewrites prompts against eval rubrics
  • Operate a personal RAG over my resume + projects + writing — Claude primary, GPT-4o + Gemini 1.5 fallback, citation-grounded responses
  • All systems Dockerized on AWS EC2 + Nginx with CI/CD, structured logging, and error budgets — production-grade, not demoware

Aug 2021 – Jul 2023

Hyderabad, India

Software Engineer — NLP & Full-Stack — Wipro Ltd.

  • Built an AI code-review assistant on Python + OpenAI Codex — cut PR cycle time 25%, adopted by 30+ engineers across 4 product teams
  • Designed an NLP ticket-classification pipeline (BERT-based) hitting 91% accuracy on 5,000+ tickets/month — reduced manual triage by 40%
  • Embedded an LLM chatbot into 3 enterprise portals serving 10,000+ users — deflected ~40% of repetitive support queries
  • Built a reusable React component library adopted across 3 products, accelerating delivery by 30%
  • Led a jQuery → React SPA migration cutting page load 45%; tuned MySQL schemas and indexes across tables of 5M+ records (~55% query speedup)

Education

M.S. Computer Science — New Jersey Institute of Technology

2025

Machine Learning · Deep Learning · NLP · Human-Computer Interaction · Database Systems

B.E. Computer Science — JNTU, Hyderabad, India

Issue 05

Full stack, full context.

Cross-functional capabilities enabling end-to-end delivery of modern applications and AI systems.

01

LLM Systems

GPT-4 / 4o / o-seriesClaude (Sonnet / Opus)Gemini 1.5 ProLlama 3 / Mistral (vLLM)Prompt EngineeringStructured Outputs · JSON ModeStreaming · SSEFunction / Tool CallingToken & Cost OptimizationFine-tuning · LoRA / QLoRA

02

Retrieval & RAG

LangChainLlamaIndexChromaDBPineconeFAISSpgvector / PostgresHybrid (BM25 + Dense)Cross-Encoder Re-rankingSemantic / Recursive ChunkingDocument Pipelines · Unstructured.io

03

Agents & Orchestration

ReAct / Plan-and-ExecuteLangGraphMulti-Agent SystemsTool Use · Function CallingMemory · Short / Long TermMCP / FastMCP ServersAnthropic & OpenAI SDKsDSPyWorkflow State Machines

04

Evaluation, Safety & Observability

RAGASLangSmith · LangFuseWeights & BiasesLLM-as-JudgeHallucination DetectionGuardrails · NeMo / LlamaGuardPrompt Injection DefensePII RedactionRed-Teaming · Eval Harnesses

05

Frontend & AI UX

React 18 · Next.js 14TypeScriptVue.js 3 · PiniaTailwind CSSStreaming Chat InterfacesServer Components · RSCWCAG 2.1 / AAFramer MotionDesign Systems · Figma

06

Backend, Data & Infrastructure

Python · FastAPI · PydanticNode.js · ExpressPostgreSQL · RedisMongoDBDocker · Kubernetes (basics)AWS · EC2 / S3 / Lambda / ECSVercel · RenderGitHub Actions · CI/CDNginx · Observability (CloudWatch)

Issue 06

Thinking in public.

Lessons learned from shipping production code, optimizing AI pipelines, and solving hard technical problems.

01

Open Source

Building an Open-Source MCP Server for AI Coding IDEs

How VIBETTER provides structured, source-grounded intelligence to Claude Code and Cursor using FastMCP 3.0.

Mar 20267 min read
02

RAG Architecture

Scaling Enterprise RAG: Lessons from 10,000+ Pages

How systematic evaluation and recursive chunking reduced hallucinations by 60% in a production knowledge retrieval system.

Jan 20269 min read
03

LLM Systems

Multi-Agent Prompt Optimization with LangGraph

Orchestrating specialized agents to iteratively refine, evaluate, and test LLM prompts.

Nov 20258 min read
04

AI Agents

Building a Multi-Step Reasoning Agent with Tool Use

Architecting an autonomous ReAct agent with GPT-4 function calling, dual memory, and self-correction loops.

Aug 20256 min read
All writing

Issue 07

Have a problem worth solving?

I’m open to full-time AI engineering roles — retrieval, agents, LLM evaluation, or the full stack around them. I reply within a day.

Primary channel
rondlanbr@gmail.com
Write an email
Open to full-time rolesJersey City, NJOPT / F-1 STEM Eligible
LinkedIn/in/neerajbhargav
GitHub/neerajbhargav