OptiHashi — ChatGPT for Business Optimization

Editor's note. This page distills two source documents: the OptiHashi Strategic Intelligence Briefing (February 19, 2026), which covers patentability, competitive landscape, and market opportunity; and the earlier Optimal: AI-Native Optimization as a Service vision doc from the lumina5 roadmap. Dollar amounts, ARR figures, and pricing tables from the sources have been generalized per the strategy archive's scope rules — see commercial materials for specifics. Public industry and competitor figures are preserved where they carry the argument.

What it is

OptiHashi is a conversational optimization engine. A user describes a business problem in plain English — “I have 30 engineers, 45 projects, 12 weeks, some projects depend on others, maximize revenue while keeping morale high” — and OptiHashi formulates the problem mathematically, races multiple solvers (classical, quantum-inspired, and quantum) in parallel, and returns a visual answer with natural-language explanation in the same conversational session.

It is the first product we know of that closes the full loop: plain English → formal QUBO/MIP formulation → multi-solver execution → interactive Gantt/allocation visualization → one-click action — all without requiring the user to write a single line of mathematical optimization.

OptiHashi is the centerpiece of B4M's quantum work. It is the demo that drove the IonQ engagement, it is the first product where the V3 platform architecture gets to flex its full agentic stack, and — per Erik's 2026-05-18 tier reorganization — it now sits as a Lab-tier research bet rather than an Engine line. The IonQ engagement itself continues through Bucket 1 (Engine) and Bucket 2 (Forge); OptiHashi is the speculative wedge alongside them, with patent upside.

Market thesis

The pitch in one line

Gurobi and CPLEX are incredible tools — for the roughly 350,000 people on Earth who can use them. OptiHashi makes optimization accessible to every business professional. ChatGPT for business optimization. And it's time.

Why now

The pieces have existed for decades, but they have never been assembled into a single conversational product. Four converging forces make this the moment:

1. LLMs got good enough to formulate. Stanford's OptiMUS (2023–2025), Microsoft Research's open-source OptiMind (Nov 2025), and a wave of academic papers (LLMOPT, OptimAI, OR-LLM-Agent) have validated that natural-language-to-formulation works. The Decision Optimization CoPilot Manifesto (Feb 2024) literally describes the product OptiHashi has built — and explicitly notes that nobody has shipped it.
2. Quantum hardware is real. IonQ, D-Wave, and Rigetti have shipped systems where QUBO-shaped problems can be meaningfully accelerated for specific structures. Hybrid classical+quantum is now consensus best practice across D-Wave, IBM, Microsoft, and NVIDIA.
3. The vibe-coding precedent. Replit, Lovable, and Cursor have proven that “natural language in, working artifact out” can take a previously-expert-only domain (writing code) and expand its addressable population by orders of magnitude — reaching multi-billion-dollar valuations in record time. Optimization is the exact same shape of problem.
4. Nobody has shipped it. This is the critical finding from our February 2026 competitive landscape work. The integration is the product.

The size of the opportunity

The traditional optimization software market (Gurobi, CPLEX, FICO Xpress, etc.) is roughly $1–2bn in annual licensing. That serves ~350K operations research professionals and data scientists worldwide.

The TAM for democratized optimization is the union of several adjacent markets:

Adjacent market	~2030 estimate (public industry figures)
Operations optimization software	$15–45bn
Quantum computing (optimization-dominant)	$7–20bn
AI decision intelligence	$35–50bn
Supply chain optimization	$48–63bn
Manufacturing execution / scheduling	$26–30bn

Roughly 90% of businesses still use spreadsheets for optimization-class problems, and 88% of business spreadsheets contain errors. McKinsey estimates half a trillion dollars in supply chain savings alone from AI-driven optimization. The ratio of “people with optimization problems” to “people who can solve them mathematically” is roughly 1000:1.

This is a material market opportunity. See commercial materials for our specific revenue plan.

The white space

The competitive map from the February 2026 briefing:

Category	Threat level	Why
Solver engines (Gurobi, CPLEX, OR-Tools)	Low	Infrastructure, not product. No NL interface. Potential integration partners.
NL-to-formulation research (Stanford OptiMUS, MS OptiMind)	Medium	Validates the thesis. Open-source components we can use.
Developer optimization platforms (Nextmv, Timefold)	Low–Medium	Developer-focused, not business-user-focused.
Decision intelligence (Aera, Diwo)	Low	Recommendations, not math optimization.
Quantum compute (D-Wave, IonQ)	None	Partners, not competitors.
Conversational BI (ThoughtSpot, Hex)	Low	Analytics, not OR.
Vibe-coding platforms (Replit, Lovable, Cursor)	None	Different domain. Useful as valuation comps.

Nobody — academic or commercial — has shipped the full conversational loop as a product. That is the wedge.

How it works

At a strategic (not technical) level, OptiHashi has five conceptual layers:

1. Natural language interface. Voice, chat, Slack, email, API. The user describes the problem the way they'd describe it to a coworker.
2. Problem classification agent. An LLM-driven agent extracts decision variables, objectives, and constraints; classifies the problem (portfolio allocation, job-shop scheduling, vehicle routing, assignment, stochastic programming…); and generates a validated mathematical model.
3. Solver selection & racing. OptiHashi spawns multiple solvers in parallel — classical MIP (Gurobi, CPLEX), constraint programming (OR-Tools CP-SAT), and quantum / quantum-inspired (QAOA, Iterative-QAOA via IonQ) — tuned to the problem's size and structure. They race; the best solution wins.
4. Result interpreter. Instead of x = [1,0,1,1,0,...] obj=18432.7, the user gets “I selected these 12 projects. Project #7 won because it has the highest revenue-to-effort ratio. If you wanted #14 instead, here's the cost.” Trade-offs are surfaced as conversation.
5. Actionability. One-click execution — create the JIRA tickets, assign the GitHub issues, post the Slack update, schedule the kickoffs.

The technical novelty — the __uiSideEffect protocol

The patentable core sits between layers 3 and 5. When an LLM tool call returns a result, OptiHashi's wrapper inspects the payload for a __uiSideEffect: true sentinel marker, validates the payload against a whitelist of effect types, persists it to the quest document in MongoDB, and a client-side dispatcher reads the persisted effect and dispatches it to Zustand stores. This is a decoupled, extensible protocol for LLM tool calls to trigger arbitrary client-side state mutations — the bridge that lets a conversational agent reach into the UI and actually render the Gantt chart, not just talk about it.

Combined with parallel Web-Worker-based solver racing and the two-phase quantum_formulate → quantum_schedule tool chain, it is the specific, concrete technical mechanism that turns a chat session into an interactive optimization workbench.

The QUBO dogfood example

Our own portfolio optimization — what becomes of B4M's 2026 roadmap when we feed it as a QUBO problem — is the canonical demo. It is also the personal-origin story: 25 years of weighted-sum portfolio ranking turned out to be QUBO all along.

Customer engagement — IonQ

OptiHashi is what walked into the room at IonQ. The demo — “describe an optimization problem in English, watch a quantum-classical race resolve it, get the answer back as a Gantt chart” — reframed B4M from “yet another AI agent shop” into “the team that makes quantum advantage usable.”

IonQ's response was to ask for two buckets of applied AI work alongside continued OptiHashi development. The SOW shape:

• Bucket 1 — OptiHashi as the wedge. The conversational optimization product itself, with quantum solvers wired into the racing engine. This is the demo asset, the patent asset, and the joint-marketing asset.
• Bucket 2 — broader applied AI / agentic work. Cognitive workbench capabilities (Deep Agents, RAG over technical corpora, multi-agent orchestration) applied to IonQ's internal and customer-facing needs.

See commercial materials for the specifics of the engagement; the strategic point is that OptiHashi was the proof-of-life that opened the door to a much larger relationship.

Patent posture

A provisional patent is in progress. The February 2026 legal analysis concluded the system is patentable with careful claim drafting. The shape:

• Alice Step 1 risk is real. Optimization and scheduling are inherently mathematical, and a 2025 Federal Circuit case (Recentive Analytics v. Fox) made “generic ML applied to a new domain” claims weaker.
• Alice Step 2 is where OptiHashi wins. The __uiSideEffect sentinel protocol is a specific, concrete technical mechanism that improves how conversational AI systems interact with client-side application state — an improvement to computer functionality, which is exactly what the November 2025 Ex parte Desjardins precedent and the December 2025 MPEP revision tell examiners to look for in AI/ML claims.
• Prior art risk is low. No identified prior art combines NL formulation + sentinel-based UI side effects + parallel multi-solver racing + interactive visualization + quantum backend in one product.

What's protectable

Three independent claims anchor the strategy:

1. The UI side-effect protocol — the sentinel-marker-plus-dispatcher mechanism.
2. The concierge flow — the end-to-end “NL → validated schema → multi-solver race → visualization” method.
3. The multi-solver racing architecture — the system that spawns parallel solvers, aggregates streaming progress, ranks by objective function, and renders the comparative result.

The combination is novel; the individual components are not. Our defensible position is the integration, plus the side-effect protocol as a discrete invention with no identified prior art.

Strategic implications for Bike4Mind

OptiHashi is not just one product in the catalog. It is a forcing function for the rest of the platform.

• It validates the V3 architecture. The agentic stack — Deep Agents, tool-calling LLMs, sentinel-based UI control, persistent quest state — was built to do exactly this kind of work. OptiHashi is V3's coming-out demo. See B4M V3 Blueprint.
• It is a Lab-tier research wedge with patent upside. In our Engine / Forge / Lab / Backlog framework, Lab products are speculative bets with learning-per-week as the primary metric. OptiHashi is the cleanest Lab-tier research bet we have, because the patent posture, the IonQ relationship, and the dogfood story all reinforce each other — even though it's not booking revenue itself. (Pre-2026-05-18 framing tracked OptiHashi as Engine; that was the wrong tier — it's a Lab play with patent and partnership upside, not a revenue line.)
• It pulls quantum into the core, not the fringe. Without OptiHashi, “quantum” is a buzzword we sprinkle on slides. With it, quantum is a solver in a race, picked when it wins, ignored when it loses — honest, technically defensible, and consistent with our integrity bar.
• It is a recruiting magnet. “Help us build the ChatGPT of optimization with IonQ as our first customer” is a meaningfully better pitch than “another AI agent platform.”
• It de-risks the IP story for fundraising / strategics. A filed provisional with novel claims around a working product changes the conversation with both investors and potential acquirers.

What's next

The shape of the work, in dependency order (no calendar):

• Phase A — Provisional filed, dogfood live. File the provisional. Run the QUBO portfolio optimization against our own roadmap. Make it the standing internal demo.
• Phase B — IonQ joint demo hardened. Quantum solver wired into the racing engine. Public-quality demo asset. Joint-marketing-ready.
• Phase C — Multi-problem coverage. Beyond portfolio allocation: scheduling, vehicle routing, assignment, budget allocation, stochastic programming. The classification agent gets sharper as the problem space widens.
• Phase D — Real-time integration. B4M Pi style: connect to Slack, JIRA, GitHub, ERPs — continuous re-optimization as data changes, not one-shot solves.
• Phase E — Platform / marketplace. Templates, enterprise data integration (ERP, WMS, CRM), scheduled and automated optimization runs. The “operating system for business decisions” framing.

Open questions worth holding

• Pricing model. Per-solve, per-seat, per-workspace, or hybrid? The right answer depends on the first few customer pilots; we should resist locking in early.
• Solver economics. Gurobi licensing at scale, IonQ quantum-time costs, the open-source fallback path — the cost-of-goods story has to make sense before we promise margin.
• Vertical wedge. The biggest TAMs are supply chain and manufacturing, but the easiest first sale may be engineering-team portfolio allocation (where the dogfood story is the demo). Resolve sequencing on the back of the first two pilots.
• Patent breadth. How aggressively do we file beyond the provisional? Continuation strategy depends on the funding picture.
• Brand. “OptiHashi” is the working name. Whether it survives a serious commercial push is a marketing decision, not a strategy decision — but worth flagging.

---

Cross-references: Strategy update — 2026-04-04 (IonQ engagement) · B4M V3 Blueprint · 2026 Roadmap Strawman · Mission, Values, Principles.