Jcentrix
Website:
jcentrix.com
Job details:
We are building vertical AI systems that actually work in production — not demos, not wrappers, not “AI features.”
Role Description
This is a full-time, on-site role for an AI Architect located in Noida. Responsibilities include designing and implementing AI architectures, developing scalable and efficient software solutions, and integrating AI systems.
Your first responsibility is to design and ship one complete AI engine that becomes the architectural template for the rest of the team.
This includes:
- Designing data ingestion pipelines from customer systems (CRM, events, voice, etc.)
- Building LLM orchestration pipelines (Claude + OpenAI) with structured outputs
- Implementing retrieval systems (RAG) using Postgres + pgvector
- Creating a production-grade evaluation harness
- Designing cost models (per prediction + at scale)
- Setting up:
- Prompt versioning
- Drift detection
- Output quality monitoring
- Writing clear architecture documentation that other engineers can replicate
Once the pattern is established, you will:
- Design new engines across multiple verticals
- Pair on complex builds
- Review and guide other engineers
Our Stack:
LLMs: Anthropic Claude (primary), OpenAI GPT (secondary)
- Orchestration: Structured outputs (Pydantic/Zod), tool calling
- Retrieval: Postgres + pgvector (default)
- Backend: Python (FastAPI) or Node.js
- Frontend (context): Next.js, TypeScript, Tailwind
- Database: Supabase (Postgres, multi-tenant with RLS)
- Evaluation: Custom or tools like Langfuse / Braintrust
- Infra: AWS / Vercel, Inngest (queues), Clerk (auth)
We prefer simple, direct architectures over heavy frameworks.
You Should have
1. Production LLM Experience
You have shipped LLM-powered systems to production in the last 18 months:
- Real users
- Real scale
- Real failures you’ve debugged
2. Strong Architecture Thinking
You default to:
- RAG
- Structured outputs
- Prompt engineering
(Not fine-tuning — unless clearly justified)
3. Evaluation Expertise
You have built evaluation systems, not just prompts:
- Labeled datasets
- Accuracy + hallucination tracking
- Drift detection
- Iteration based on eval results
4.Leadership Ability
You can:
- Design systems others can replicate
- Review code rigorously
- Guide mid-level engineers
- Treat quality as a gating function
Nice to Have- Voice pipelines (Whisper, Deepgram, AssemblyAI)
- Multi-tenant SaaS experience (RLS, isolation)
- Experience in vertical AI (legal, healthcare, fintech, etc.)
- Strong technical writing
- Comfortable working across multiple products
Click on Apply to know more.