scroll or click empty space to release
Blanc Quant Services · founded by Jean Blanc
tl;dr — founder, Blanc Quant Services · senior quantitative engineer, 15+ years across utilities, industrial R&D, standards and energy analytics. On my own time: deterministic C++ trading infrastructure whose CI fails if latency regresses, and GOTHAM, the AI second brain rendering this page.
The reliability infrastructure layer for mission-critical AI systems. AI and quantitative systems that survive production — Blanc Quant Services designs and deploys governed AI, automation, analytics, and performance infrastructure for organizations operating in complex, high-consequence environments. 20+ systems shipped. The edge: seeing connections others miss, synthesizing across domains, designing frameworks that make both operational. One standard: the system must work, measure itself, and produce evidence.
Proof of capability: you're inside GOTHAM, my personal AI operating system — a privacy-scrubbed export of the real thing. 5,208 notes, indexed and answerable by six agents on seven models, plus an 18-sage council for the hard calls. Real structure, real counts, content stripped; it rebuilds itself at 3 AM every morning and scores itself nightly. Click a cluster to fly in.
01 — the galaxy
Every professional connection, mapped as a constellation.
Entry nodes — where every new thought lands before the graph files it.
Life, family, the questions behind everything else.
Strategy, moves, the long game.
Systems, tools, and how they break.
The companies, tools and places my notes keep mentioning.
02 — how i can help
Every engagement begins the same way: a 30-minute assessment call, then a defined problem, fixed scope, measurable acceptance criteria, and a clear evidence package.
For regulated teams where governance is
the system that blocked the thing — controls that live in PDFs, AI approvals measured
in months, and zero usable traces to hand an examiner.
You get policy that runs at the moment of
use: eval gates and kill criteria enforced in CI, and evidence that writes itself —
so the governed path is the fastest one.
For teams whose models, data, agents and
workflows don't talk to each other — pilots that never reach production, AI output nobody
can audit.
You get one governed production system:
retrieval, routing, permissions, evaluation and observability, with an evidence trail for
every consequential action.
For infrastructure-heavy organizations —
utilities, industrial, operations — where decisions live in spreadsheets, tribal knowledge
and manual reporting.
You get decision tools, analytics pipelines,
knowledge systems and automated workflows that measure themselves.
For trading desks and compliance teams where
performance claims are anecdotes.
You get deterministic replay, p50–p999 latency
gates enforced in CI, digest verification and retained evidence bundles — via
BQL Engine.
03 — the numbers
where the galaxy comes from
GOTHAM is one expression of that philosophy. What you're flying through is a privacy-scrubbed export of the real system — genuine structure and counts, regenerated from the live corpus, with the content stripped. GOTHAM itself is locally hosted on my own hardware: loopback-bound, kill-switched, encrypted at rest. Nothing on this page holds a live connection to my private data, models, accounts, or production environment.
selected results — full record on the founder page
04 — case study · gotham
Genesis · Oracle · Tars · Hugo · Aegis · Max · armed with 100+ licensed skills · six agents × seven models — 117,649 possible fleet configurations
hover any node — agents, sages, stages — for its roletap a node for its role — swipe sideways to pan the diagram
A browser-side simulation of GOTHAM's routing rules — no live connection to the real system: name an agent to address it directly, and anything unclaimed defaults to MAX. In the real fleet, saying GOTHAM fans one prompt across six agents in parallel over four model families, then MAX synthesises. Every skill is licensed to exactly one agent and enforced at the call site — an agent cannot perform an operation it does not hold.
05 — blanc quant services
Three names, one engineer — said once, plainly: Blanc Quant Services (this site) is the consultancy — the engagements in section 02. Blanc Quant Systems is the product company — quantitative proof and AI-governance infrastructure, starting with BQL Engine. GOTHAM is the internal system both are proven on, built to the same evidence standard.
The product: Blanc Quant Systems — BQL Engine 2.0 — deterministic proof infrastructure for market systems. It replays market workloads, verifies identical behaviour, and blocks latency regressions in CI — catching what no ordinary test can: a change that silently alters book state or tail latency. Performance claims in trading are anecdotes, a benchmark someone ran once on a machine no one still has. This turns them into a contract CI enforces on every commit.
canonicalise inputs → replay deterministically → gate everything = same digest + evidence bundle
Deterministic C++20 replay for limit-order-book workloads, p50/p95/p99/p999 gates enforced in CI, canonical digest verification, and retained evidence bundles (manifest · bench.jsonl · metrics · summary) a regulator or a new engineer can re-derive months later. The bundle is the unit of proof — artifacts a regulator can re-run. Built for HFT and prop desks where “it feels faster” is not an answer, and for compliance teams who need the receipt.
06 — software portfolio · 20 builds
Every build below measures itself and leaves evidence.
07 — the audit
Ten dimensions, self-scored. Lowest score shown first. Total 85/100. The three lowest scores from the first audit, organisation, memory and resilience, were fixed and re-scored.
the kill certificate — governance, as evidence
The same standard governs AI-proposed trading strategies. LLMs propose, tests decide: every candidate runs a gauntlet of statistical gates, and each verdict writes a signed evidence digest as it happens. This is a real verdict: a strategy the gates rejected. It is the artifact a buyer shows their examiner.
One gate failed, so the strategy died — five passing gates bought it nothing. That is the point: a control that can't kill anything isn't a control. Offer 01 installs this discipline on your stack.
08 — the ask
Everything on this page runs for real, every day — built by the same engineer who ships a deterministic C++ order-book engine where CI rejects any p99 breach. The fleet gets the same discipline: security gates on every egress, failover that survives rate limits, receipts for every claim. If your firm needs AI that survives production, bring Blanc Quant onto the problem.