AI Architect

OVERVIEW:

infrastructure

AI Architect

not

systems + architecture + applied AI

You will work closely with:

• Product Owner / Architect (founder-led)
• Backend Engineers
• AI/ML Engineers
• DevOps
• Frontend (for AI-to-UX contracts)

think in systems

CORE RESPONSIBILITIES:

AI System Architecture:

• Design and own the 

  end-to-end AI architecture

  , including:
• Kernel / Orchestrator layer
• Engine-based execution model
• LLM abstraction & routing layer
• long-term memory
• Context Bus & context lifecycle
• Artifact generation & storage model
• Define 

  clear contracts

   between:
• API → Kernel
• Kernel → Engines
• Engines → Artifacts
• Context Builder → LLM Client

Orchestration Design:

• Architect the 

  Kernel

  :
• Accept structured requests (JSON DSL)
• Decide what engine(s) to run, in what order
• Manage execution state, retries, failures
• Define 

  engine selection logic

   (static + dynamic):
• Rule-based
• Context-based
• Project-based
• Design async workflows (job queues, schedulers).

Framework & Workflow Design:

• Design a 

  pluggable Engine architecture

  :
• BA Engine (PRD/BRD/FRD)
• Tech Architecture Engine
• Estimation Engine
• Code/Flow Generators
• Define:
• Engine input/output schemas
• Engine chaining & dependencies
• Versioning strategy for engines
• Ensure engines are 

  stateless

  , reproducible, and auditable.

Context and RAG:

• Own the 

  context window strategy

   end-to-end:
• Context artifact selection
• RAG inclusion/exclusion rules
• Compression / summarization strategies
• Design multi-step context strategies:
• Artifact-first context
• Selective RAG injection
• Optional artifact summarization
• Decide 

  what lives in context vs memory vs artifacts

  .

LLM Routing:

• Design an 

  LLM Client Layer

   that:
• Supports multiple providers (OpenAI, Anthropic, etc.)
• Allows model switching per task
• Handles token budgets, retries, fallbacks
• Define:
• Which models are used for which engine
• Cost/performance tradeoffs
• Guardrails and response validation

Memory:

• Architect 

  Memory

• Long-term project memory
• Artifact relationships
• Version history
• Define artifact lifecycle:
• Creation
• Updates
• Dependencies
• De-duplication
• Work closely with DB design (Postgres + pgvector).

Guardrails & Safety:

• Implement guardrails for:
• Hallucination control
• Schema validation
• Output consistency
• Ensure deterministic outputs where required (especially for docs).
• Define confidence scoring, validation hooks, and rejection logic.
  

MUST HAVE (NON NEGOTIABLE):

• 8+ years experience in 

  AI / ML systems

• Proven experience designing 

  LLM-based platforms

• Strong understanding of:
• LLM orchestration patterns
• RAG systems (vector DBs, embeddings, retrieval strategies)
• Context window limitations & mitigation
• Hands-on experience with:
• Python (FastAPI preferred)
• JSON-based AI workflows
• Async processing (queues, schedulers)

Strongly Preferred:

• Experience building 

  multi-tenant AI systems

• Experience with:
• PostgreSQL + pgvector
• Redis
• S3-like artifact storage
• Familiarity with:
• Prompt templating systems
• Output schema enforcement (Pydantic / JSON Schema)
• Prior work on:
• AI copilots
• AI product builders
• Agent frameworks (but not blindly following them)

Mindset we care about (Very Important):

• Thinks in 

  systems, not demos

• Hates prompt spaghetti
• Understands tradeoffs between:
• Intelligence vs reliability
• Flexibility vs determinism
• Comfortable saying:

What this role is not:

• Not a research-only role
• Not a prompt engineer role
• Not a chatbot builder
• Not a data scientist role

AI infrastructure architecture