LLM-Powered Applications
Production chat, coaching, and copilot experiences on OpenAI, Anthropic, Llama, and DeepSeek — with streaming, latency budgets, and failover built in from day one
Build Your AI Product →I build LLM products people actually use
Founding engineer at COROS AI. I built its 24/7 coaching platform from zero — the backend, the retrieval and memory, the guardrails, and the iOS and Android apps. I take on one outside project at a time.
At COROS full-time · one freelance slot open this quarter
Live demo
Don’t read about my work. Ask it.
This is a real LLM, grounded in my resume with the same retrieval and guardrail patterns I ship for clients. Ask how I cut latency 78%, how the failover works, or whether I’m free next quarter.
ask-maaz — live LLM, grounded in my resume
$ ask anything about my work — answers stream from a real model, grounded in my resume.
Services
RAG & Knowledge Retrieval
Semantic search and retrieval pipelines over your docs, catalogs, or conversations: chunking, embeddings, Pinecone, re-ranking, and answers with citations
Ground Your AI →AI Agents & Automation
Multi-agent systems that plan, call tools, and finish the job — turning messy inputs like meeting notes into structured, actionable output
Automate the Work →Full-Stack Web Apps
Next.js, React, TypeScript, and Node from concept to deployment — the product UI, APIs, auth, and billing around your AI core
Ship Your App →iOS & Android Apps
React Native apps shipped end-to-end — CI/CD, store submission, and the full release lifecycle — sharing one backend with your web product
Go Mobile →Voice & Conversational AI
Voice clones, speech interfaces, and conversational flows with ElevenLabs — engineered for real conversations, not demos
Give Your Product a Voice →LLM Evals & Guardrails
Eval pipelines for retrieval and memory quality, plus guardrails that catch harmful or off-policy responses before they reach users
Make It Safe →Cloud, DevOps & Scaling
GCP and Azure infrastructure, Docker, microservices, and request-based scaling that keeps the bill down as usage grows
Scale with Confidence →AI Product & Architecture Advisory
Architecture reviews, model and vendor selection, and a pragmatic roadmap for founders deciding what to build first
Get a Second Opinion →Founding engineer, for hire
“The same standards I hold at COROS, I bring to every client build.”
Production AI That Actually Ships
LLM products engineered for real users — latency, reliability, and safety handled, not hand-waved.
As founding engineer at COROS AI I built a 24/7 LLM coaching platform from zero: the backend, the retrieval and memory systems, the guardrails, and the iOS and Android apps. Now I bring that same end-to-end ownership to founders who need an AI product shipped, not prototyped.
No More Demo-ware
Prototypes are easy; production is where AI products die. I design for latency budgets, streaming, and graceful failure so the experience holds up under real traffic.
78%
Cut in AI response latency at COROS (27–32s → 6–10s)
Reliability You Can Trust
Zero-downtime LLM failover across OpenAI, DeepSeek, Together AI, and Llama. Guardrails that block harmful responses before delivery. Evals that prove memory and retrieval actually work.
4
LLM providers behind one failover layer
End-to-End Ownership
Backend, retrieval, mobile, CI/CD, and cloud cost — one engineer who has shipped the whole stack, so you don't have to coordinate five contractors.
2
App stores shipped, iOS + Android, from zero
The Process
- 1Discovery Call30 min
- 2Scope & Architecture2–3 days
- 3Build & Demoweekly
- 4Launchyour infra
- 5Ongoing Support30 days
100+
concurrent users served
35+
U.S. cities scaled
95%
fewer double-bookings
2
LLM products in production
Engaged When and How You Need
Flexible engagement models tailored to your stage, budget, and timeline.
Advisory
2–5 hours/week
Architecture reviews, model selection, and unblocking your team
Fractional
1–3 days/week
An embedded senior AI engineer without the full-time hire
Project
Fixed scope
MVP to launch with a clear roadmap and weekly demos
Full-time
5 days a week
Dedicated founding-engineer partnership for the long haul
Case Studies
A showcase of my work across LLM systems, retrieval, full-stack product, and mobile — all of it running in production for real users.
COROS AI Coaching Platform
24/7 LLM coaching platform built from zero: semantic memory, four-provider failover, AI guardrails, and a production ElevenLabs voice clone
NextBook AI
LLM book recommendation system in daily use at NED University's library — semantic search over the full catalog with Pinecone and Llama
PinPoint AI
Turns messy meeting notes into organized task lists with assignees and deadlines, extracted in real time by Llama models
PRIV Marketplace
On-demand beauty marketplace scaled across 35+ U.S. cities — real-time availability, WebSocket notifications, and a mobile-first booking UI
COROS Mobile App
iOS and Android app for the COROS platform shipped end-to-end — CI/CD, store submission, and the full release lifecycle
From Prototype to Production
Three systems I’ve shipped to production, drawn the way I’d whiteboard them for you.
LLM routing with zero-downtime failover
One router in front of OpenAI, DeepSeek, Together AI, and Llama. Health checks and latency budgets decide the primary; a provider outage swaps the backup in mid-stream with no developer in the loop.
RAG that cites its sources
Chunk → embed → Pinecone → re-rank → answer with receipts. Retrieval scoring is measured by an eval pipeline, not vibes — that's where the 78% latency cut at COROS came from.
Semantic memory with decay
Topic-isolated memory lanes with time-based decay and recency-weighted recall, so one conversation never pollutes the next and the model remembers what still matters.
AI / ML
RAG & Multi-RAGMulti-Agent SystemsRAG Eval PipelinesMemory System EvalsPrompt & Context EngineeringLLM Failover & Routing
Frontend
ReactNext.jsTypeScriptTailwindFramer Motion
Backend
Node.jsPythonWebSocketsPostgreSQLMongoDBRedis
Cloud / Infra
GCPAzureDockerCI/CDMicroservices
Tools
Claude CodeCodexCursor
Early-stage and budget-conscious? Covered.
You don't need a five-person AI team to ship a production LLM product. One founding-level engineer who has done it before gets you to launch faster, and cheaper.
Learn more about my services →Impact
78%faster AI responses
27s → 6sLLM latency at COROS
4providers, zero-downtime failover
100+concurrent coaching sessions
35+U.S. cities scaled at PRIV
−95%double-bookings
+50%mobile bookings
−40%missed appointments
2app stores shipped from zero
1university library running NextBook AI
#1Data Insights Hackathon 2024
78%faster AI responses
27s → 6sLLM latency at COROS
4providers, zero-downtime failover
100+concurrent coaching sessions
35+U.S. cities scaled at PRIV
−95%double-bookings
+50%mobile bookings
−40%missed appointments
2app stores shipped from zero
1university library running NextBook AI
#1Data Insights Hackathon 2024
Got an AI product that needs to ship?
Tell me what you're building and where it's stuck. I'll tell you honestly whether I'm the right person for it — and if I'm not, who might be.
Book a 30-minute call