MissionControl — Proposal Workspace

Overview

We're building ChangeCurve.ai — an AI-native enterprise transformation platform that helps organizations navigate complex organizational change at scale. We're at a critical inflection point: our prompt-based generation pipeline works, but we need it to evolve into a true reasoning system with structured outputs, modular orchestration, and retrieval-backed intelligence.

We're looking for a senior AI architect who can own this evolution end-to-end, from prompt architecture to RAG integration to production deployment.

What We Have Already Built

A working LLM pipeline using GPT-4 and Claude for transformation intelligence generation
Domain-specific prompt templates covering 12 transformation domains
Supabase backend for data storage and user state management
Activity frameworks and structured output schemas (partial)
A React/Next.js frontend connected to a FastAPI middleware layer

What We Need Help With

Redesigning our prompt architecture for modularity, composability, and reliability
Building a RAG layer on top of our existing Supabase infrastructure — vector store selection, chunking strategy, retrieval optimization
Moving from free-form generation to schema-bound outputs that activity frameworks can consume programmatically
Implementing reasoning chains and multi-step orchestration for complex transformation scenarios
Establishing evaluation infrastructure — automated quality checks, regression testing for prompt changes

Ideal Experience

Production RAG deployments — you've shipped it, not just prototyped it
Deep familiarity with LangChain, LlamaIndex, or equivalent orchestration frameworks
Python/FastAPI backend development
Vector databases: Pinecone, ChromaDB, Weaviate, or similar
Experience with structured output and function-calling patterns across major LLM providers
Comfort with agentic patterns — multi-step reasoning, tool use, memory systems
Strong technical communication — we move fast and you'll be working closely with our CTO

Engagement

This is a contract role, 20–30 hours/week initially, with potential to scale. We're looking for someone who can start within the week. Fully remote, async-friendly, strong preference for overlap with US East Coast hours.

Back to Feed Proposal Workspace

claude-sonnet-4-6

~$0.024

Context

Prompt

Draft

Submit

Profile

RAG & Knowledge Systems

Active context profile

Portfolio Matches

2 projects matched by tag overlap

ContactOS Voice AI Platform

rag llm voice-ai fastapi

DataStudios · 2024

4 tags

Dev Agent — Agentic Triage System

agentic llm python multi-agent

DataStudios · 2024

3 tags

Summaries will be injected into the prompt as portfolio evidence

Global Keywords Matched

RAG LLM Agentic Python Vector DB LangChain

Assembled Prompt

SYSTEM: You are a senior proposal writer for DataStudios.AI, a boutique AI consulting firm specializing in RAG systems, LLM application development, and enterprise AI architecture. Write proposals that are direct, technically credible, and grounded in real production experience. Avoid fluff. Lead with specific evidence. PROFILE CONTEXT (RAG & Knowledge Systems): DataStudios specializes in RAG architecture, LLM application development, and vector database design (Pinecone, ChromaDB, mem0). Diego Sanz leads all AI architecture engagements. The firm has shipped production RAG systems handling concurrent enterprise workloads, voice AI platforms with retrieval-backed knowledge bases, and agentic triage systems with multi-step reasoning chains. PORTFOLIO EVIDENCE: [ContactOS] DataStudios built ContactOS, a production voice AI platform for enterprise clients. The system uses a modular prompt orchestration layer, RAG-backed knowledge retrieval over a Pinecone vector store, and structured reasoning chains. Handles 300+ concurrent users with consistent output quality. Stack: FastAPI, Python, GPT-4 + Claude, Pinecone. [Dev Agent] Diego led development of an agentic triage system (DevAgent) for DataStudios internal use. Two-phase architecture: fast classification (~6s) followed by deep investigation (~2.4 min). Uses mem0 for persistent memory across sessions and LangChain for orchestration. Full MCP server integration. JOB: Senior AI Prompt Engineer / Enterprise AI Architect CLIENT: ChangeCurve.ai DESCRIPTION: We are building an AI-native enterprise transformation platform called ChangeCurve.ai. We need to evolve our prompt-based pipeline into a true reasoning system with modular orchestration, RAG integration, and schema-bound outputs... KEYWORDS: RAG, LLM, Python, Agentic, Vector DB, LangChain, FastAPI INSTRUCTIONS: Generate a concise, specific proposal (200-300 words) that leads with the most relevant portfolio evidence. Reference ChangeCurve.ai by name. Be direct. End with a clear next step.

1,840 tokens input 4 sections Assembled from profile + 2 portfolio matches + job

AI Proposal Draft

claude-sonnet-4-6 · 847 tokens output

Your ChangeCurve.ai architecture is at exactly the inflection point where prompt-based generation needs to evolve into a true reasoning system — and I've built that transition before.

At DataStudios, I recently architected ContactOS, a production voice AI platform that required exactly the kind of modular prompt orchestration you're describing: domain-specific prompt managers, RAG-backed knowledge retrieval, and structured reasoning chains that handle ambiguity without hallucinating confidence. The system handles 300+ concurrent enterprise users with consistent output quality — the kind of reliability that comes from architecture, not just prompt quality.

For ChangeCurve.ai specifically, I'd approach your transformation intelligence system in three layers:

Prompt architecture — modularizing your existing domain prompts into composable units with explicit input/output contracts, then building orchestration logic that selects and chains them based on transformation stage and user context.
RAG integration — your Supabase infrastructure is a strong starting point. I'd layer a vector store (Pinecone or ChromaDB) on top for retrieval-augmented reasoning, letting the system pull relevant transformation patterns and heuristics rather than generating from scratch each time.
Structured output — moving from free-form generation to schema-bound outputs that your activity frameworks can consume programmatically. This is where enterprise reliability lives.

I work in Python/FastAPI, have production RAG deployments, and can start immediately. Happy to review your existing prompt architecture on a scoping call.

Generated in 8.2s

2 portfolio projects were injected as context — ContactOS Voice AI Platform, Dev Agent Triage System Show ›

Input: 1,840 tok Output: 847 tok Est. cost: $0.024