All of AI, delivered to production.
Strategy, agents, generative AI, retrieval, computer vision, analytics, platforms and integration. For teams that need a system they can run and defend, not a demo. From a few hours of expert help to a multi-year build.
What we do
Six practice areas covering the whole AI lifecycle. Most engagements combine several.
Which problems AI can solve for you now, what each will cost, and what is realistically achievable. Roadmaps, build-versus-buy decisions, model and vendor selection, and an honest reality check on the ROI.
Task agents, multi-agent orchestration, tool calling, AI work plans, drafting and summarization, headless AI APIs for other systems to call. Natural-language access to large databases and telemetry, with text-to-SQL that learns from corrections. Any model, any provider, with routing and failover. See it on 372 billion rows.
RAG that survives large corpora, conflict-resolution rules when documents disagree, hybrid search, reranking, knowledge graphs, and TruthLock verification so answers carry a grade and a signed record.
Schematics, drawings, photos and scans. Symbol detection, image tagging, page classification, OCR, and ingestion into knowledge bases with zero data loss. Detection, ingestion and tagging in detail.
Multi-tenant platforms with LLM routing across providers, circuit breakers, guardrails, observability and cost controls. Prompt and preset governance, evaluation harnesses, audit trails, and enablement so your team can own what we build. Security reviews against OWASP ASVS before go-live.
AI wired into the systems you already run: Salesforce and Agentforce, ServiceNow, Microsoft 365, Google Workspace, AWS, ERP and CRM, through APIs, managed packages, MCP and connectors.
Specialists on call when needed
Iron Rod Systems leads every engagement with a single point of accountability. When a job needs more hands or deeper specialization, we bring in specialists we have delivered with before, and manage them under one contract.
That includes Salesforce Field Service, Data Cloud and Agentforce implementation teams, ServiceNow developers, deep AWS infrastructure and networking, CI/CD and DevOps engineering, front-end and mobile builders, and data engineering.
- Salesforce
- Field Service, Data Cloud, Agentforce, Service and Revenue Cloud implementation partners
- ServiceNow
- Platform developers and integration specialists
- AWS
- Deep infrastructure, networking, security and cost engineering
- CI/CD and DevOps
- Pipelines, infrastructure as code, release and observability engineering
- Build teams
- Front-end, mobile, data engineering and QA as scope requires
- Accountability
- One contract, one lead, one set of traceable results
Engagements of every size
A few hours of expert help is a real engagement. So is a multi-month platform build. Most clients start small and grow the scope once they see results.
A fix, a component, a review
- A custom Agentforce response component with confidence and citations
- A headless AI API demo for a partner's app
- An architecture or security review of your AI stack
- A website with an AI assistant for a small business
A proof on your real data
- Symbol detection on a real drawing set, scored against your own counts
- A plain-English analytics agent over one of your databases, with charts and briefs
- A cited-answer support API over your manuals, served into Salesforce, ServiceNow or your own portal
- TruthLock wired into an existing assistant
A platform or a product line
- Document ingestion deployed in your own cloud with governed retrieval
- A multi-tenant AI application platform with failover, guardrails and observability
- A managed Salesforce package across many AI providers
- A dedicated Iron Rod AI platform instance, run for you
Fixed-price for small and medium work when the scope is clear. Time and materials or milestone pricing for larger builds. Every size gets the same rule: results you can trace.
Track record
Work delivered by Iron Rod Systems and, before it, by its founder. Client names are shared on request and with permission.
Invented Elle, visual intelligence that reads technical schematics, wiring diagrams and blueprints inside Salesforce Agentforce.
Patent pending. Named in Digital Journal's "AI Tools Fueling Startup Growth This Year." Shipped as a managed package with reference-in-place architecture, so documentation stays where it lives.
Created the product that applies generative AI to any Salesforce object: case summaries, AI work plans, knowledge article drafts, cross-sell suggestions.
Grew from v3.5 to a v6.4 managed package with a headless Ask Elevate API, per-org spend limits, token cost tracking, and named credentials for more than 15 AI and vector providers.
Engineering document ingestion deployed in the client's own AWS account, and AI contract review against a preferred-terms rule library.
Vision models classify each page (schematic, wiring diagram, blueprint, civil drawing) before OCR, chunking and indexing into a governed knowledge base with retry, recovery and integrity checks.
An AI operations center over 372 billion rows of telemetry, with natural-language SQL that learns from its own fixes.
Became the first prototype deployment of Breathing Dragon AI.
A multi-tenant AI application platform with installable apps for analytics, contract compliance, spec compliance and document Q&A, built in four months.
LLM routing with failover across Claude, GPT and Gemini, tenant isolation on every call, guardrails, and CloudWatch alerting on every tenant.
An RF-DETR vision and image detection layer built into RenderDraw's working takeoff and quoting platform, so every count is cross-checked against the drawing.
Detector trained per drawing set on auto-generated labels, legend-template fallback, symbol readiness grades, active learning from review. First proven job: 104 of 106 symbols on unseen regions of a 19-sheet MEP set, 98%, after 74 minutes of training on a rented vast.ai GPU.
"Huge credit to Jeff Hunt. His engineering intuition and relentless drive have been instrumental in making this possible."Lou Simon, Chief AI Officer, Uptima. More on LinkedIn
How we build
The rules are the same whether it is a consulting deliverable or a product release.
Anything that has to be right, a count, a price, a deadline, a policy, comes from the source and a written rule, not from the model's judgment. The model helps you read and write; it never gets the final say on a fact.
A figure links to the page, row or box it came from. A claim links to the source and tier it rested on. A record is signed, so what was true on the day stays provable later.
Every agent has a declared scope: which tools, which data, which actions. What it may display, quote, send or do is decided by a reliance policy, with a person in the loop wherever the policy says so.
Tenant isolation on every call, least-privilege access, secrets outside the code, PII redaction, audit logs, and an OWASP review before go-live. Your data stays in your account and is never used to train anything.
An evaluation set built from your own ground truth, run on every change. If a result cannot be measured, it is not a result.
Presets, agents, taxonomies and rules live in configuration with drift detection. The next job is a config change, not a rewrite.
Failover chains, circuit breakers and health checks are built in from the first commit. When something fails, you know.
Documented, deployed in your cloud or ours, with your team enabled to run it. No lock-in, no hostage data, no surprise at handover.
How an engagement runs
Short, measurable phases. You see working software early.
A working session on your real problem and the data, systems or documents behind it. We agree on the one measurement that defines success.
A system that runs on your real data, documents, systems or workflow, scored against your own ground truth. Days to a few weeks.
Production hardening: tenancy, security, monitoring, failover, documentation. Deployed in your cloud or ours.
Your team owns it, or we operate it. Either way, every number stays traceable.