All clients

OOTI

BTP | ERP

An AI agent that lets construction teams query financial data in plain language. 92% accuracy at < $0.10 per call.

OOTI builds ERP software for construction companies. Their users. project managers, finance teams, site supervisors. needed access to complex financial data but lacked the technical skills to write queries or navigate dense dashboards.

Rakam built an AI Data Agent that transforms natural language questions into precise data queries, with streaming responses, an insights tool that generates code on the fly, and a feedback loop for continuous improvement. all at a cost of < $0.10 per interaction.

« OOTI's AI agent transforms how construction teams interact with their financial data. natural language replaces complex queries. »

O

OOTI

Client partnership

92%

Tool accuracy

97%

Faithfulness score

< $0.10

Cost per call

100%

Response consistency

Business

Business Impact

Construction teams now access their financial data through natural language queries instead of complex dashboards. This removes the technical barrier that previously required finance or IT staff to pull reports, saving hours per week across project sites.

At < $0.10 per interaction, the AI agent is cost-efficient enough to deploy across all user tiers. The 92% tool accuracy and 97% faithfulness scores mean users trust the answers, reducing back-and-forth vérification with finance teams.

Product

What We Built

AI Data Agent

Core conversational agent that translates natural language questions into precise ERP data queries. Understands construction-specific terminology. projects, budgets, margins, timelines. and returns structured, contextual answers.

Insights Tool (Code Generation)

Dynamic code generation tool that creates custom analytical queries on the fly. When standard tools aren't enough, the agent writes and executes code to derive insights from complex, multi-table financial data.

Streaming & Feedback System

Real-time streaming responses for a responsive UX, combined with a user feedback system that captures satisfaction signals to continuously improve agent accuracy and relevance.

JWT Authentication & Security

Secure authentication layer using JWT tokens to ensure each user only accesses their authorized data scope. Role-based access control aligned with OOTI's existing permission model.

Technical

Technical Architecture

Built on the Rakam Systems framework, the agent uses async LLM calls for low-latency streaming. The insights tool dynamically generates Python code for complex analytical queries that go beyond predefined tools.

MLflow tracks all agent interactions, measuring tool accuracy, faithfulness, and cost per call. Sentry provides error monitoring, while Kubernetes OVH ensures reliable deployment and scaling.

# Stack

Rakam Systems (agent framework)

Async LLM (streaming)

MLflow (observability)

Sentry (error monitoring)

Kubernetes OVH (orchestration)

JWT (authentication)

# Performance

Tool accuracy: 92%

Faithfulness: 97%

Cost: < $0.10/call

Consistency: 100%

Move from AI hype to the real thing

Leave us a message, our team will get back to you within 24 hours.