Saravanan Kannan · Founder, IgniteEdge AI

Agentic Quantitative Trading, Built on Blackwell Metal

Autonomous LLM agents for trade intelligence, sub-millisecond risk, and 14-state order execution — running live on a dual-node NVIDIA Grace-Blackwell GB10 cluster.

Not a chatbot bolted onto a dashboard. Seven production services, on-prem GPUs, 200 Gbps RDMA fabric, and a CME-calibrated margin engine — all speaking open protocols (llms.txt, MCP, x402).

View Resume Explore the Platform
17
Production Containers
235B
Parameter LLM Inference
<0.3%
SPAN 2 Margin MAPE vs CME
200 Gbps
RDMA/RoCE Fabric

Live Platform Services

Every service below is live in production and exposed through a Cloudflare Tunnel. Click any hostname to hit the real endpoint.

Spark

LIVE
Agentic Trade Intelligence · FastAPI + Streamlit
  • DSPy DAG + Gemini for trade-idea generation
  • SABR smile, risk reversals, term structure analytics
  • ChromaDB-backed RAG knowledge base
  • /v1/market-insights, /v1/portfolio-recommendations
  • x402 micropayment gating · MCP discovery
spark.igniteedge.ai →

Mercurius

LIVE
Sub-Millisecond Agentic Risk Firewall
  • 28-station sequencer pipeline
  • RegT · Portfolio Margin (TIMS) · SPAN processors
  • Gate A (LLM semantic, Ollama qwen3) + Gate B (math)
  • 50+ strategy matchers · 4 option pricing models
  • Java NaN-arithmetic semantics preserved in Python
mercurius.igniteedge.ai →

Agentic OMS

LIVE
HMAC-Signed Order Management
  • 5-agent pipeline: Spark → Mercurius → OMS → SOR → Settlement
  • 14-state validated order lifecycle
  • HMAC-SHA256 signed inter-agent ledger
  • Graph-based smart order routing (NetworkX)
  • 177 passing tests · Redis-backed order store
oms.igniteedge.ai →

SPAN 2

LIVE
CME Margin Engine (Clean-Room Rebuild)
  • x·HVaR + (1−x)·Stress + Liquidity + Concentration
  • Pod-based margining with cross-pod offsets
  • Two-stage scipy.optimize calibration
  • 60+ calibration tests · <0.3% MAPE vs CME
  • Black-76, BAW, BS2002, SABR smile, EVT/GPD
span2.igniteedge.ai →

SSETF

LIVE
Single Stock ETF HFT Platform
  • 47 ETFs · 17 underlyings · IBKR via ib-insync
  • HMM regime detection (micro) + LLM macro
  • Auction signal (NYSE tick 588 imbalance)
  • Rebalance flow engine · 8 TRS formulas
  • Dual LLM: Qwen3-80B Instruct + Thinking
ssetf.igniteedge.ai →

Binary Options

LIVE
CBOE XSPBW / QSB Agentic Trading
  • Dashboard tab on Spark · launches June 2026
  • Binary + vertical pricing (XSPBX settlement)
  • Agent orchestrator on Qwen3.6-35B-A3B
  • System 1 (fast) + System 2 (thinking) split
  • Human-in-the-loop proposal approval
via spark.igniteedge.ai →

PNR Dashboard

LIVE
P&L Replication & Attribution
  • Cross-asset P&L decomposition
  • Driver attribution (delta, vega, theta, rate)
  • Companion to UnifiedRBM margin engine
internal dashboard

Platform Capabilities

Machine-readable, agent-native, and pay-per-call — the platform is built for the protocols the next generation of financial AI agents will speak.

llms.txt Discovery

Every service exposes /llms.txt and /llms-full.txt — machine-readable descriptions of endpoints, schemas, and capabilities for AI agent discovery.

all services

MCP Protocol

Model Context Protocol endpoints let AI agents discover and invoke tools programmatically, with typed schemas and structured results.

Spark · Mercurius · SPAN 2

x402 Micropayments

Pay-per-call API access gated by x402 on Base Sepolia. No API keys — just on-chain USDC. Verified with live on-chain transactions.

Spark · Mercurius

Agentic Architecture

Services operate as autonomous agents — Spark generates, Mercurius validates, OMS executes. HMAC-signed ledgers make every step cryptographically auditable.

platform-wide

On-Prem GPU Inference

Dual-node NVIDIA Grace-Blackwell GB10 cluster. TensorRT-LLM + vLLM. NVFP4 / FP8 quantization. 235B-parameter models in production.

LocalInfra

Multi-Broker Execution

major-broker TraderAPI + IBKR (ib-insync with fully unattended IBC v3.23). Single broker-abstraction layer across the stack.

Spark · OMS · SSETF

About Saravanan

I'm a senior engineering leader with 20+ years at the intersection of institutional trading infrastructure and frontier AI systems. I build the thing end-to-end: the LLM agents, the risk math, and the GPU metal they run on.

Currently leading the AI-native modernization of Organization's Risk Monitor (P0 initiative — 10× analyst triage speedup), while independently architecting UnifiedQuant, Mercurius (sub-ms agentic risk firewall), Agentic OMS, UnifiedRBM (cross-asset RBM), and LocalInfra (dual Grace-Blackwell cluster running 235B-parameter inference at 15.4 tok/s).

Fluent in CME SPAN 2 mathematics, agentic LLM orchestration (DSPy DAGs, MCP, x402), Blackwell-class GPU optimization (NVFP4, TensorRT-LLM, RDMA), and — most importantly — the strategy / factory / visitor patterns that keep a JVM-era trading core safe during Python translation.

Read the Full Resume → Get in Touch →