01
Full-Stack Developer
SupportIQ
AI Customer Support Platform
Next.jsReact 19TypeScript
2026Case study
NR* — 001 · Jersey City, NJOpen to Forward Deployed roles
● Issue 01 · an introduction
Hi, I’m Neeraj.
I take ambiguous problems from zero to production AI systems — owned end-to-end, model to interface to infrastructure, with zero handoffs.
The engineer you hand the ambiguous problem to. I run the technical discovery, decompose it, and ship the whole stack — retrieval, agents, evaluation, the UI, and the infrastructure under it. Currently the sole AI engineer at Landair Advisors, running five production systems in NYC commercial real estate, where a hallucination costs real money. Based in Jersey City, NJ, targeting Forward Deployed Engineer roles.
Scroll · 02 · About
Issue 02
One engineer, the entire AI stack.
One person carrying a feature from the data layer to the pixels — so nothing gets lost in a handoff, because there isn't one.
“I’m the full vertical — retrieval, agents, evaluation, streaming UI, and the AWS pipeline that ships it. Fewer tickets. Faster iteration.”
Currently the sole AI engineer at Landair Advisors (NYC commercial real estate) — five production systems, owned and operated end-to-end. Open-source on the side: VIBETTER, Context Refinery, OmniContext, and the CRE Intelligence Platform. Previously Innovcentric (RAG, agents, ~60% hallucination cut) and Wipro NLP. Targeting Forward Deployed Engineer roles.
i.
Own it end-to-end.
From discovery to deployment to the 2 a.m. page — no handoffs, no excuses.
ii.
Sit with the problem first.
Technical discovery before code. Framing the right system is the hard part.
iii.
Ship systems, not demos.
Deployed, monitored, behind CI/CD — or it doesn't count.
iv.
Evaluation is the product.
If you can't measure it, you can't put it in front of a customer.
Outside the terminal
- Photography
- Gym
- Gaming
- Reading
- Travel
- Anime
- Motorcycles
- Cars
- Building things end-to-end
Issue 03
Personal & Open-Source.
Alongside my enterprise work, I independently build these tools and platforms to stay hands-on with the rapidly evolving agentic AI stack. Drag or scroll to explore.
Showing card 1 of 5
Issue 04
History & stack.
Three roles, one habit — take a fuzzy AI problem, ship a system you can measure, then keep shipping. Here's the toolkit behind it.
Experience
Applied AI Engineer — Owns 5 Production AI Systems
2026 – PresentLandair Advisors · New York, NY · On-site
- Applied AI Engineer at a NYC commercial real-estate brokerage — inherited and own five production AI systems end-to-end on NYC property data (ACRIS, PLUTO, DOB, NY DOS)
- Skiptracer — LLC-chain owner resolution via a NY DOS → ACRIS → IDI waterfall with confidence scoring; measured 93.3% owner-resolution accuracy over 1,121 records. Python · Supabase · Claude Haiku
- CompScope — Supabase database of 9,391+ comparable sales with PLUTO + ACRIS asset-type classification, powering broker deal analysis
AI Software Engineer — Full Lifecycle, Zero Handoffs
Nov 2025 – Jun 2026Innovcentric LLC · Remote · USA
- Owned the complete Gen AI strategy and implementation lifecycle for enterprise customer-experience platforms — technical discovery with product owners and operations managers, problem decomposition from ambiguous business needs into production AI architectures, pilot validation before full rollout, and post-deployment optimization — one engineer, model to interface to infrastructure, zero handoffs
- Built a production-grade RAG pipeline (LangChain + ChromaDB + FastAPI + Pydantic) over 10,000+ pages of support documentation — PDF/DOCX ingestion, recursive chunking (512-token windows, 50-token overlap), OpenAI embeddings, and metadata filtering — delivering sub-2s semantic search for 100+ support agents and replacing manual document lookup entirely; benchmarked 4 chunking strategies and 3 embedding models to tune retrieval
- Deployed an autonomous ReAct agent on GPT-4 function calling with MCP-style tool-use chains — knowledge retrieval, account-context lookup, multi-step reasoning, and sub-agents for specialized tasks — running 5–10 tool calls per query with self-correction loops, dual memory (conversation buffer + vector store), and Human-in-the-Loop escalation for billing-sensitive edge cases
Software Engineer — NLP, AI Tools & Full-Stack
Aug 2021 – Jul 2023Wipro Ltd. · Hyderabad, India
- Built and deployed an AI-powered code-review assistant (Python + OpenAI Codex) with custom prompt engineering for logic-error and code-smell detection — cutting PR merge cycles 25% and adopted by 30+ engineers
- Designed an NLP ticket-routing pipeline (Scikit-learn + HuggingFace BERT, Pandas, NumPy) hitting 91% accuracy on 5,000+ tickets/month and reducing manual triage 40%; delivered a conversational AI chatbot with intent detection, FAQ automation, conversation memory, and escalation logic across 3 enterprise portals — 40% fewer repetitive queries for 10,000+ users
- Built and maintained enterprise React applications at 99.5% production SLA — interactive components, form validation, Chart.js dashboards, and Lighthouse/code-splitting performance work; created a reusable component library that accelerated delivery ~30% and led a jQuery → React SPA migration for 45% faster page loads
Stack
8 domains, end to end — the tools I reach for to take an AI problem from prototype to production.
Forward-Deployed Delivery
01End-to-End Gen AI OwnershipZero-to-One Production DeploymentTechnical Discovery & ScopingProblem DecompositionAmbiguity NavigationPilot → ProductionStakeholder CommunicationWhite-Glove Onboarding · L1–L2Runbooks · Office HoursPost-Deployment Optimization
LLM Systems
02GPT-4 / 4oClaude Sonnet 4.5 / OpusGemini (2M context)Llama 3 / Mistral (Ollama)Prompt Engineering · VersioningStructured Outputs · JSON ModeStreaming · SSEFunction / Tool CallingSFT · LoRA / QLoRAToken & Cost Optimization
Retrieval & RAG
03LangChainLlamaIndexChromaDBPinecone · FAISSpgvector · WeaviateHybrid (BM25 + Dense + RRF)Cross-Encoder Re-rankingSemantic / Recursive ChunkingAgentic RAGDocument Ingestion Pipelines
Agents & Orchestration
04ReAct · Plan-and-ExecuteLangGraphMulti-Agent SystemsMCP · FastMCP 3.0Sub-Agents · Tool RegistryMemory (buffer + vector)Human-in-the-LoopDSPyasyncio · Event-DrivenWorkflow State Machines
Evaluation & Observability
05RAGAS · DeepEvalLangSmith · AgentOpsArize PhoenixLLM-as-JudgeHallucination DetectionFaithfulness ScoringDrift MonitoringSafety GuardrailsPrompt Injection DefenseOpenTelemetry
Machine Learning & NLP
06HuggingFace TransformersScikit-learnTensorFlow · PyTorchText Classification · NERSummarizationTransfer LearningCNNs · ResNetGrad-CAMPandas · NumPyMLOps
Frontend & AI UX
07React 18/19 · Next.js 16TypeScriptVue.js 3 · Vue FlowTailwind CSS v4Streaming Chat UIsServer Components · RSCWCAG 2.1 AAFramer MotionTauri v2 · LeafletFigma · Lighthouse
Backend & Infrastructure
08Python · FastAPI · PydanticNode.js · NestJSJava · GraphQL · WebSocketsPostgreSQL · Prisma · SupabaseMongoDB · RedisDocker · Docker ComposeAWS (EC2 / S3 / Lambda / ECS)Vercel · GitHub ActionsNginx · MicroservicesHealth Checks · Structured Logging
Education
M.S. Computer Science — New Jersey Institute of Technology (NJIT)2023 – Dec 2025
B.Tech. Computer Science & Engineering — JNTUH — Hyderabad, India2018 – 2022
OPT / STEM extension eligibleIssue 05
Thinking out loud.
Notes from actually building this stuff — what worked, what I'd do differently, and the decisions that don't fit in a README.
Showing card 1 of 3
Issue 06
Have a problem worth solving?
Open to full-time AI roles, and always up for a good problem. Drop me a line — I usually reply same day.