Edgeforge Tech

Senior engineering and applied AI, built for production.

Boutique software engineering practice serving enterprise clients across North America. Ten-plus years building production systems where failure is not an option.

Email meLinkedIn

years building production systems: 10+
peak bets per minute sustained: 100,000+
uptime SLAs during major events: 100%
licensed jurisdictions served: 75+

Selected work

AI-Native Learning Management Platform

US training organization

Challenge: Replace legacy training delivery with a multi-tenant, AI-native LMS for compliance-heavy technical content, serving multiple customer organizations from one platform.
Built: Multi-tenant SaaS LMS with RAG over course content (PostgreSQL + pgvector semantic search), document chunking pipeline, SCORM/xAPI and LTI 1.3 interoperability, role-based dashboards, offline-capable PWA. React/Next.js + TypeScript, Python services, Azure OpenAI.
Result: Platform scoped and delivered to production on a phased six-month plan within a six-figure enterprise engagement; Phase 2 expansion now underway.

Multi-Tenant Fleet Operations Portal

commercial laundry services operator

Challenge: Institutional customers had no self-serve view of the equipment fleets on their campuses. Uptime questions, service escalations, and SLA reporting all routed through account managers.
Built: Multi-tenant SaaS portal over the equipment vendor fleet API: executive uptime and SLA dashboards, plus facilities-team drill-down to the individual machine. NestJS and Next.js on Amazon ECS Fargate, PostgreSQL with per-tenant scoping in the service layer and a row-level-security backstop, Cognito auth with tenant-scoped token claims, infrastructure in Terraform with keyless OIDC deploys.
Result: Complete portal delivered against a 240-hour engagement, with two large university systems onboarded on live fleet data across 150+ campus locations.

Field Operations Platform with AI Assistant

regional mutual insurance carrier

Challenge: Field representatives logged agency visits on paper and in scattered notes. Leadership had no consistent read on agency performance, and underwriting-sensitive figures needed tight internal access control.
Built: One platform, three surfaces sharing auth, data, and infrastructure: a digital field-visit report with offline drafts and sync, an agency-performance dashboard, and an AI assistant on Amazon Bedrock whose tools are bound to the same authorization scopes as the rest of the app. NestJS API and queue worker plus Next.js web on Amazon ECS Fargate, PostgreSQL through RDS Proxy with IAM authentication, row and column level authorization in the service layer.
Result: Three surfaces delivered inside the first two months after kickoff: reps capture visits offline in the field, leadership reads agency performance off the same data, and the assistant answers questions strictly inside the scopes each user already has.

LLM-Powered Bid & Project Intelligence

heavy-civil construction contractor

Challenge: Decades of bid and project history locked in unstructured documents; leadership had no structured way to analyze past performance by project type.
Built: LLM-powered document processing and classification pipeline across ~420 historical projects: automated extraction, a project-type taxonomy, and a queryable analysis layer.
Result: First structured, firm-wide view of the complete project and bid history, delivered as a focused block-of-time engagement.

Investment Adviser Intranet and Data Layer

registered investment adviser

Challenge: A registered adviser with no in-house technology team. Fund accounting, investor records, and bank data lived in three separate vendor systems, and every internal report was assembled by hand.
Built: Internal intranet with Microsoft Entra ID single sign-on and four permission groups, an hourly cache that pulls all three vendor APIs behind one internal REST envelope so no browser or dashboard ever holds a vendor credential, a self-registering dashboard loader, and an embedded Claude assistant. Python and Flask on Windows Server with IIS.
Result: Delivered against a 30-hour block of time, with dashboard authoring left in-house: a new dashboard is one HTML file dropped in a folder, no code change and no release.

Global Real-Time Betting Platform

OpenBet, six years as senior engineer

Scale: 100,000+ peak bets per minute, 100% uptime SLAs during major sporting events, 75+ licensed jurisdictions, powering top-tier operators including FanDuel, William Hill, and Ladbrokes.
Role: Front-end and back-end development across a globally distributed, high-availability betting engine; the foundation for the reliability discipline applied to every Edgeforge project.

Case studies anonymized to respect client confidentiality. Happy to discuss details on a call.

How I deliver

Discovery
Short call to understand the problem, constraints, and success criteria.
Written scope
Phased plan with explicit in and out of scope, risks, and budget.
Working software first
Usable releases early, before the full budget is committed.
Support
Production support, iteration, and a clear hand-off path.

Engagement models

Hourly

Senior development or architecture work, billed transparently against tracked hours.

Monthly block of time

Reserved senior capacity each month, the right fit for ongoing product work.

Fixed-scope build

A scoped phase with a defined deliverable, for well-bounded projects.

Core stack

Frontend

React / Next.js
TypeScript
React Native

Backend

Node.js (NestJS, Express)
Python

Data

PostgreSQL + pgvector
Prisma
Redis
MongoDB

AWS

Lambda
S3
Cognito
SQS
EventBridge
Step Functions
DynamoDB
SES
Textract
CDK

AI

Bedrock
OpenAI / Azure OpenAI
Anthropic
RAG and agent pipelines

About

I am Michael Batrakov, Principal Software Engineer and CTO of Edgeforge Tech. I spent six years at OpenBet building real-time betting platforms that handled 100,000+ peak bets per minute under 100% uptime SLAs, and the last several years leading AI and full-stack delivery for enterprise clients in learning management, financial services, insurance, and construction tech.

I work from British Columbia, Canada, with clients across North America.