Dhananjay (Deej) Raol
When a problem is complex, critical and has no template, I'm the person leaders call. Now I solve those problems with AI.
Principal, Aliud Inquisito Inc.
Claude Certified Architect, Professional and Foundations
Finance first. Then the tools changed.
For most of my career I solved hard problems with financial models, process redesign and custom systems, long before AI was in the toolkit. At DriveTest, Ontario's driver examination partnership with the Ministry of Transportation, I led finance for 83 locations and built the labour and bid models behind an $850M re-bid.
From 2021 I built business models and financial architecture in clean energy and advanced materials: a vanadium flow battery subsidiary for Largo, rooftop solar and storage economics for MobiSmart, solar to hydrogen strategy for One Power and a graphene venture of my own that reached a working pilot.
Then the tools changed, so I retrained. Most strategy consultants can't architect the system that executes the plan, and most AI engineers can't tie it back to the P&L, the contract or the regulator. I do both.
- 2004 to 2019
- Serco Canada, DriveTest OntarioSenior Strategy Analyst, then Manager Strategy and Planning, then Finance Lead (Finance Director under the MTO contract). The public private partnership that runs Ontario's driver examinations.
- 2021 on
- Principal, Aliud Inquisito Inc.Business models and financial architecture for Largo Clean Energy (grid storage), MobiSmart (solar and battery storage) and One Power USA (green hydrogen)
- 2024
- Founder, graphene commercialisation ventureWith Bharat Nanoparticles, India. From lab science to a working pilot.
- 2024
- AI client work through Aliud InquisitoRibbon and UR Channel
- 2026
- Claude Certified ArchitectFoundations (July) and Professional (September), Anthropic
Watch this first.
Draft This is a draft, to be cleaned up. Everything below it is the long version.
Where the work is now.
- Ribbon Plc, Gibraltar (fintech)Jun 2024Started with the business and financial model for a new financial services company. Later, with the CEO, I designed an AI native client acquisition model for 100,000+ accounts: Dify, voice agents, RAG and LLM orchestration with decision nodes. Now I advise on AI governance ahead of the company's Canadian launch, so observability and traceability are in the platform before it is built.
- UR Channel CommunicationFeb 2025I design n8n and LLM orchestrated workflows that connect systems which never talked to each other, route work and validate data, so staff stop re-keying it. I delivered an email, calendar and CRM knowledge pipeline that is now in production.
Architect and product owner. I design the system, direct AI coding agents through the build and own every decision in it.
A memory that lives on your own machine.
The_Borg_DB is a local-first knowledge intelligence system. It turns email, meetings, tickets and documents into structured facts in SQLite plus a LanceDB vector store, with four retrieval modes. While the machine sits idle, an 18 stage pipeline farms what it has collected for patterns, and devices share what they find through the hive.
On a single NVIDIA DGX Spark with local Gemma models it processed 866 real support tickets and emails in about two hours with no failures, and its self-correcting data quality layer removed 12 to 18 percent junk entities. No data left the premises.
Read the case studyFrom inbox to a knowledge base you can question.
KBPipeline is an email, calendar and CRM knowledge pipeline. It ingests Gmail and Microsoft accounts over OAuth, strips the signatures, quoted replies and noise, classifies and extracts what is left with language models, and routes it into searchable knowledge bases. It has 187 automated tests and is deployed in production at a client, where what it surfaced between the email and ticket streams led them to change what data they capture.
Read the case studyA companion on the TV for people living with dementia.Built, pre-deployment
Kare4u composes personalised therapeutic programs (orientation, reminiscence and calm) from a family's photos, music and cloned voices, and plays them on the patient's TV. A Raspberry Pi 5 with a Hailo accelerator detects agitation on the device. The patient never touches a screen. The calm program follows the Iso-Principle: music starts at the patient's energy level and slows over 25 to 30 minutes.
Six client apps, an event driven API, a 58 table data model, and consent and audit built in. I'm open to a partner to take it into care settings.
Read the case studyAgents that know when they aren't needed.
- BORG_ROUTEAn agent routing and orchestration layer. It breaks a request into a dependency graph, sends each step to the cheapest capable model and composes the answer. Governance text survives every compression step word for word, and every piece of scratch state has an owner. 256 automated tests.
- AgentArchitectA source grounded agent design generator. Give it a task and it returns a runtime neutral specification with termination criteria, guardrails tied to the mechanism that enforces them and an evaluation plan, or it tells you an agent isn't needed. Its dry run on four tasks returned four correct verdicts. One of its rules: verification belongs in the harness, never in the agent's own prompt.
Three problems that came without a template.
- The reconciliationTen years of transactions with the Ministry, over $800M, with deferred revenue split across 20+ fee types and three entities, and rules that changed every year. It was accepted on the first pass with zero variance and no changes, which almost never happens at that size.
- The bid nobody had a template forOntario's Ministry of Community Safety and Correctional Services needed Security Guard Testing. I built the model to run it on infrastructure DriveTest already had, so it was won as new revenue at high margin.
- The battery subsidiaryLargo's co-chair engaged me as the company moved from mining into energy storage. I built the business model and cost architecture for the vanadium flow battery subsidiary from the ground up, with storage cost targeted at parity with lithium ion, and a five year plan that gave it a path to stand alone.
Decisions written down, kill tests agreed first.
Every project carries written decisions and a reason for each, so a new engineer or a cold AI session can pick it up. I state a kill test before I build, grade claims by how well they are evidenced and log the ideas that failed. Cost limits, safety and write authority are designed in before the first line of code.
The Parallel Commit Law: produce every artifact you can derive at a build checkpoint, while the model still has the context loaded.
Reloading that context later means paying for the same tokens twice. The agent already knows the error code, the origin function and the conditions that trigger it, so the error registry entry costs almost nothing to write at that moment.
- VideoHarnessIn progressA model neutral harness for AI video production, with a rule and evidence registry that grades every design claim by confidence. It is making the film on this page.
- OpenClaw connector for Open WebUIOpen sourceStreams chat to an OpenClaw agent over its WebSocket protocol, with Ed25519 device authentication and an install guide. View on GitHub
- WritingMy Parallel Commit Law (LinkedIn article, April 2026) and BORG DB: local intelligence, not cloud (LinkedIn post, August 2026). Read them on LinkedIn
Where AI earns its keep, then the build.
Every build starts with a business case and ends with something your team can run and audit.
- AI strategy and governanceWhere AI pays back and what it costs to run, with observability and traceability designed in before the platform is built.
- Knowledge systems over your own dataEmail, calendar, CRM tickets and documents turned into a searchable knowledge base, on your hardware when the data can't leave the building.
- Workflow automation and agentsn8n and LLM orchestrated workflows that connect systems which never talked to each other, route work and validate data so staff stop re-keying it.
- Finance transformationRevenue recognition, month end, R2R and O2C, labour and bid models, built by someone who has run the close.
Contract, interim, fractional or full time. Toronto, hybrid or remote.
Let's talk.
I'm open to AI solution architect, AI strategy and finance transformation work: contract, interim or full time, in Toronto, hybrid or remote.