RPJ
I Don't Build
Wrappers. I Build
Infrastructure.
Production multi-agent systems. Custom LoRA-tuned LLMs. Distributed inference pipelines processing 10,000+ daily requests. I architect the AI infrastructure that enterprise clients trust to ship.
▹Multi-Agent Systems
LangGraph, LangChain, LlamaIndex
▹Custom LLM Training
LoRA, vLLM, KServe, MLflow
▹Cloud & DevOps
AWS, Kubernetes, Terraform, Docker
▹Production Backend
FastAPI, Neo4j, Redis, Celery
Your AI Is Held Together With
API Calls & Prayers.
Most “AI products” are thin wrappers around OpenAI. One rate limit, one policy change, one outage — and your entire system folds. I build the opposite: custom-trained models you own, multi-agent orchestration that self-heals, and infrastructure that scales without a single vendor lock-in.
openai.chat.completions.create()
Vendor-locked. No fallback. No ownership.
LoRA → vLLM → KServe → K8s → Prometheus
Custom models. Your data. Your infrastructure. Zero lock-in.
Built. Shipped. In Production.
Not concept demos. Not hackathon prototypes. These systems process real data, serve real users, and run 24/7 without supervision.
Autonomous Document Validation
Visa2fly — Production SystemLangGraph-powered multi-agent system that autonomously validates complex visa documents across global immigration rules. Seven specialized AI agents work in concert — processing, normalizing, validating, repairing, and reporting — without human intervention.
- › 95% validation accuracy
- › 100+ concurrent requests daily
- › 7 orchestrated AI agents
- › 150+ audit reports per day
Custom LLM Training & Serving
Inference Cost EliminationStop burning API credits. Domain-specific language models fine-tuned with LoRA on proprietary travel and visa datasets, served on an optimized vLLM/KServe stack with intelligent caching via LMCache. Your model, your data, your infrastructure.
Context-Aware RAG System
Neo4j Knowledge GraphRetrieval-Augmented Generation pipeline powered by a Neo4j knowledge graph storing 10,000+ validation rules and document relationships. Graph-based context retrieval delivers precise, regulation-compliant answers — not hallucinations.
- › 10,000+ validation rules indexed
- › Graph-based semantic retrieval
- › Vector + Knowledge Graph hybrid search
- › Regulatory compliance guaranteed
Ship It.
Keep It Alive.
Building is 10% of the work. The other 90% is keeping it alive under load. Automated CI/CD, real-time observability, auto-scaling infrastructure, and zero-downtime deployments — all battle-tested at scale.
Ripunjay Singh
AI Engineer & Systems Architect
Experience
Education
Recognition
Let's Build Something
That Ships.
You've seen the infrastructure. I architect custom, production-grade AI systems — multi-agent pipelines, fine-tuned LLMs, and scalable cloud infrastructure. Open to relocation and remote opportunities worldwide.