About AxonGrid
Security screening for grid operators — the full contingency set evaluated inside the dispatch block, not sampled after the fact.
Our Mission
To give grid operators complete security awareness within the dispatch interval — every credible contingency screened before the block is locked, not a sampled subset reviewed after the fact.
Transmission planning tools were built for offline studies, where an engineer can wait an hour for an answer. Control rooms cannot. We build for the second case: the same rigour, inside the time the operator actually has.
Our Vision
A grid where no dispatch decision is taken blind. Where the whole contingency space is evaluated continuously rather than spot-checked, and security screening stops being a planning-desk artefact and becomes a control-room reflex.
As inverter-based generation displaces synchronous machines, the set of credible failures grows faster than any sequential tool can screen. We think that gap is closed by parallelism and by honest solver reporting — not by asking operators to trust a number they cannot audit.
Core Values
Reliability
Infrastructure-grade software built for mission-critical operations
Intelligence
Advanced algorithms derived from decades of research and practical expertise
Candour
A solver that did not converge is reported as such. We publish residuals, not reassurance
Sovereignty
Grid telemetry stays inside the jurisdiction that owns it, on-premise or in-region
Our Foundation
The tools most control rooms rely on were designed for transmission planning, where an engineer can pose a question in the morning and read the answer after lunch. They are excellent at that. They were never meant to run inside a fifteen-minute dispatch block, and it shows: operators screen a handful of contingencies because screening all of them is not on the table.
That constraint is getting worse. As inverter-based generation displaces synchronous machines, the set of credible failures grows and the margins shrink. The honest position is that sampling a few outages and extrapolating is no longer a safe approximation of the grid’s actual exposure.
A single AC-OPF is a sparse, sequential problem — a GPU offers little over a good CPU solver and pays real overhead moving the matrix into VRAM. The advantage appears when the work is independent and repeated: an N-1 sweep is thousands of unrelated solves, which is exactly the shape a GPU is built for. We use the GPU where that holds and say so where it does not.
The second half of the problem is data. A national model is stitched together from state transmission utilities, central transmission licensees and independent producers, each with its own conventions. Cases arrive with mismatched parameters, orphaned lines and floating buses. We treat ingestion and validation as a first-class part of the product rather than a precondition someone else should have satisfied.
September 2026
Product Roadmap
The dispatch engine is in production. Q4 hardens it into something an operator can rely on without checking the working; 2027 turns the GPU loose on the full contingency set.
Pilot Stack — In Production
The dispatch engine and the data path into it: CSV-native AC studies wired into the AxonGrid platform. Contingency work at this stage is operator-driven, not automated.
- Power flow (Historical Audit) — Newton–Raphson study with CERC voltage band flags (0.97–1.03 pu)
- Single Snapshot/Block AC-OPF — GPU interior-point dispatch, LMP, thermal watchlists, audit HTML/PDF
- Multi-period AC-OPF — 96 × 15-min day-ahead with temporal coupling and generator ramp limits
- Mission Control — RTS-GMLC ingest, live-solve stepper, stability gauge, certificate export
- Advanced Scenario Lab — N-k manual contingencies via CSV line/generator toggles (operator screening)
- Zonal & interstate analytics — per-zone demand, generation, LMP, inter-zonal flows, and cross-border import/export
- Platform security — org email, MFA, session controls, audit logging, always-on GPU compute in AWS Sydney
- Data Preparation guide — documented CSV case layout for SLDC engineers
A Core Worth Trusting
Before we screen thousands of cases, one case has to be beyond argument. This quarter is about the solver telling the truth and the platform sitting next to its users.
- Honest convergence reporting — a run that exhausts its iteration budget is surfaced as unconverged, with primal and dual residuals shown, never recorded as a completed study
- Convergence diagnostics — identify the binding constraints behind an infeasible case and suggest relaxations, so operators stop hand-tuning limits to coax a solution
- Large-case solve economics — bring the 10k-bus snapshot to a reliable converged result and publish the timing alongside it
- In-country data residency — production stack relocated to AWS Mumbai (ap-south-1) for latency and jurisdictional fit
- Batched contingency kernel — the groundwork for parallel N-1: many independent solves per GPU dispatch instead of sequential re-solves
The Parallel Thesis, Delivered
This is the quarter the hardware argument has to prove itself: a complete N-1 sweep inside the time an operator actually has.
- Full automated N-1 screening — every credible branch and generator outage evaluated within one 15-minute dispatch block
- Contingency ranking and violation maps — worst-case thermal and voltage exposure ordered by severity, not dumped as a table
- Prioritised N-2 screening — targeted double-outage analysis on the pairs that N-1 flags as critical
- Published reproducible benchmark — PGLIB 10k results and wall-clock timings, open enough for an evaluator to rerun
- Intra-block revision targets for 15-minute dispatch windows
Into the Control Room
Screening is only useful where decisions are made. This quarter is about integration and governance for SLDC environments.
- SCADA / EMS adapters — scheduled CSV exchange and webhooks, not black-box GUI lock-in
- Enterprise SSO & role-based access — SAML/OIDC alongside existing MFA and org-email policy
- On-prem and state SLDC VPC deployment, with commercial AMC options
- Multi-area orchestration — coordinated runs across control zones
- Transformer tap & shunt controls in OPF — phased fidelity upgrade from the snapshot pilot
- Policy-bound re-dispatch suggestions inside a block, always with human override
Certification and Scale
Long-horizon bets, once automated screening is the trusted path.
- Grid Code-grade SCOPF sign-off — regulator-accepted N-1/N-2 certification, not screening alone
- MIP and branch-and-bound unit commitment inside the dispatch problem
- RoCoF & inertia floor analytics — frequency-stability screening layered on OPF outputs
- Distributed edge optimisation — federated solves across DER clusters without centralising telemetry
- Pan-utility coordination — cross-state interchange optimisation with sovereign data boundaries
Manual N-1/N-2 in the current pilot means operators copy a case, trip lines or generators via CSV, and re-run OPF — screening for the control room, not a Grid Code certificate. See our Data Preparation guide for the Advanced Scenario Lab.