$cd .. && cat quantum-finance/README.md

quantum finance

hybrid risk + pricing libraries

qaoa allocator, amplitude-estimated monte carlo, variational kernels. quantum where it earns its keep. classical numpy everywhere else. ships into existing quant pipelines today.

problem

$50t+ aum industry runs portfolio optimisation, derivatives pricing, and tail-risk simulation on classical monte carlo. compute spend is enormous. fat-tailed distributions are routinely undercounted because the simulation budget runs out before the tail is well sampled. that's a measurement failure, not a model failure.

wedge

three primitives, hybrid stack:

  • - qaoa · combinatorial allocation (asset selection under cardinality + sector constraints).
  • - amplitude-estimated mc · variance reduction on tail-risk + pricing sims. near-quadratic theoretical speedup over classical mc; var/cvar in seconds where classical pricing pipelines need minutes.
  • - variational kernels · time-series modelling under fat tails. classical surrogates fall over here.

hardware survey

  • - ibm heron · 133 qubits, modular, gate fidelity supports depth needed for amplitude estimation on tail-risk slices.
  • - ionq forte · trapped-ion, high two-qubit fidelity. good for shallow variational circuits.
  • - quantinuum h2 · 56 qubits, all-to-all connectivity. enables qaoa graphs that ibm/ionq topologies would need swaps to emulate.
  • - all three already support production proof-of-value on a meaningful subset of allocation + tail-risk problems.

demo · qaoa knapsack

tiny 4-bit knapsack. qaoa-style probability sampling vs brute force. watch the search distribution concentrate on the optimal bitstring. classical reference shown alongside.

step 0 / 24 · β = 0.00 · p(optimal) = 6.3%
0000
6%
0001
6%
0010
6%
0011
6%
0100
6%
0101
6%
0110
6%
0111
6%
1000
6%
1001
6%
1010
6%
1011
6%
1100
6%
1101
6%
1110
6%
1111
6%
items · a(v4,w2) b(v5,w3) c(v3,w2) d(v7,w4)cap · w ≤ 7optimal · 0101 · v=12 w=7

eval discipline

backtests w/ real frictions (slippage, transaction cost, borrowing rate). walk-forward only. no train-on-test. stability metrics over single-run alpha. the same rigor that ai evaluation needs from the inside out: leakage-safe, agent-level, replicable.

target buyer

quant funds + insurers buying compute reduction first, not quantum advantage theatre. classical-first foot in the door via amplitude-estimated mc reducing runtime on existing pipelines. ramp quantum slice as hardware scales.

roadmap

  • - q3 2026 · qaoa portfolio allocator + ae-mc for tail-var. private design partners.
  • - q4 2026 · open sdk + paper. paid hosted runtime for the slow paths.
  • - 2027+ · variational kernel library on top-of-stack as hardware fidelity grows.
  • - by 2028 · fault-tolerant hardware unlocks quantum monte carlo speedups that matter. api + library layer built before then becomes the moat.

adjacent track · finance llm eval

classical quant gives ground truth. ai is the candidate. evaluating llm finance agents against the qf primitives w/ private hold-out sets, agent-level scoring, no train-on-test. built into the same library so eval rigour is the default, not a bolt-on.