Omara Technologies
Farsight
The central governance and intelligence portal orchestrating an enterprise-grade AI data ecosystem.
Timeline
Oct 2024 — Jan 2025
Role
Fullstack AI Software Engineer
Core Stack
Go / Python / Next.js
Infrastructure
AWS / Neo4j / LangGraph
Fig 01. // GT Scoring Intelligence Dashboard
The Thesis
Transforming dark data into actionable intelligence through systematic governance.
At Omara Technologies, I spearheaded the development of high-impact platforms designed to bridge the gap between LLM-driven automation and robust enterprise infrastructure.
Farsight acts as the central brain—orchestrating Knowledge Graph generation (DocuNexus) and highly scalable human-in-the-loop (HITL) workflows through a unified documentation and management portal.
DocuNexus
Knowledge Graph Generation
Agentic AI Workflows
The "Brain" of Document Intelligence
DocuNexus is a powerful platform designed to extract, analyze, and visualize relationships within massive PDF repositories by converting them into structured Knowledge Graphs using Gemini 1.5 Pro and Neo4j.
Agentic Search Workflows
Implemented complex search agents using LangGraph, allowing for multi-step reasoning and deep document retrieval that moves beyond simple semantic search.
NL-to-Cypher Engine
Engineered a Natural Language to Cypher query engine, enabling users to query complex graph data using plain English by bridging LLM intent with Neo4j schema awareness.
Relational Discovery
Automated the extraction of "first-class relationships," making it possible to discover hidden links across disparate document sets (Legal, Insurance, Healthcare).
Cloud-Native Sync
Built robust ingestion pipelines for AWS S3 and Google Drive, ensuring seamless, scalable document synchronization for enterprise clients.
Labeling Platform
High-Concurrency Go Backend
Enterprise Workflow Orchestration
The Infrastructure for High-Precision Data
To power high-stakes AI models, I built the Enterprise Labelling Platform—a comprehensive system for managing large-scale document annotation tasks with a focus on consensus, accuracy, and throughput.
Scalable Task Distribution
Developed a custom Go-based engine supporting Parallel and Series distribution strategies. This allows for both high-throughput parallel cycles and strict sequential consensus reviews.
Consensus & Arbitration
Implemented an automated consensus layer that identifies agreement using distance metrics, with a specialized Arbitration Hub for Subject Matter Expert (SME) conflict resolution.
Backend
Go (Gin / GORM)
Database
PostgreSQL JSONB
Frontend
Next.js 15 / shadcn
Auth
AWS Cognito RBAC
Reliability Engineering
Data-Centric
over Model-Centric
I pioneered a "Data-Centric" approach at Omara, realizing that the biggest gains in AI performance came from improving the quality of the training data through the proprietary **Ground Truth (GT) Scoring Framework**.
01 // Reliability
Utilized statistical methods like **Cohen’s Kappa** to ensure that annotator agreement was scientifically rigorous, not accidental.
02 // Precision
Built systems to calculate field-level precision and recall, ensuring mission-critical data (Financials/Legal) had near-zero error rates.
03 // Loops
Developed "closed-loop" feedback systems where model edge cases identified in review were fed back into training sets.
Ending Notes
Beyond Engineering:
The AI-First Mandate.
My time at Omara was defined by a single, uncompromising philosophy: Intelligence is the primary citizen.
In most organizations, AI is a layer added at the end. For me, Farsight was the proof that AI must be the foundation. This "AI-First" approach meant that every line of Go in the backend and every Neo4j schema was architected specifically to be consumed and enhanced by autonomous agents.
Philosophy // 01
"Don't build features for users; build intelligence engines that empower them."
Philosophy // 02
"Data is noise until it's governed by Ground Truth."
Farsight stands as a testament to what happens when you stop treating AI as a tool and start treating it as the architect of the system itself.