6+ years engineering high-throughput distributed telemetry pipelines, userspace security SDKs in Rust & C++, full-stack TypeScript / Next.js / PostgreSQL platforms, autonomous MCP multi-agent workflows, and automated financial settlement ledgers.
Zero mockups. Every repository and platform is live in production, tested, and cryptographically verified.
Production interactive command & control (C2) demonstration simulating a 20Hz spacecraft downlink stream (Launch Live Sandbox ↗).
Demonstrates real-time telemetry decoding, sub-50ms browser Web Crypto HMAC-SHA256 frame verification, automated anomaly triage (HV bus voltage drops <82V, reaction wheel RPM jitter >5,800 RPM, payload thermal runaway >65°C), and batch compaction into single-roundtrip PostgreSQL unnest() columnar arrays.
Direct architectural alignment with K2 Space (Ground Software), Astranis (Flight / Network), and Mach Industries (Deployment & Flight Systems).
Architected and deployed from zero to production (gg-loop-tournament.vercel.app).
Features a multi-tenant B2B Studio Telemetry Ingest Engine (/api/v1/ingest, /api/v1/flags) powered by single-roundtrip PostgreSQL unnest() bulk inserts, sub-50ms WebSocket state streams, and automated anti-cheat anomaly rules (flagging impossible input rates >25 CPS and metronomic macro variance <2ms).
Backed by an ACID-compliant double-entry PostgreSQL financial ledger, Stripe Connect multi-party payouts, and idempotent cryptographic webhook replay protection.
Engineered a lightweight Ring-3 userspace behavioral telemetry and anti-cheat crate in Rust (github.com/djjrip/anti-cheat-sdk).
Implements Win32 process memory enumeration (K32GetModuleFileNameExW), SHA-256 binary integrity hashing, sub-millisecond click/input timing entropy analysis, and HMAC-SHA256 cryptographic event signatures.
Designed and deployed production Model Context Protocol (MCP) tool servers and multi-agent LLM pipelines integrating Claude, OpenAI, and AWS Bedrock.
Features dynamic JSON schema validation, deterministic tool execution, automated telemetry incident triage, and pgvector RAG retrieval.
Managed high-volume cash settlement operations across national ATM networks at NCR Atleos. Audited daily transaction settlement flows, configured ATM deposit and interchange fee structures, and executed monthly cash payouts and account reconciliations in coordination with senior IT analysts.
Detailed engineering deep-dives covering high-throughput telemetry ingest, cryptographic tamper-resistance, autonomous MCP agents, and double-entry FinTech ledgers.
The Bottleneck: Naive ORM implementations execute discrete INSERT statements per packet. At 20Hz across multiple subsystems, this produces hundreds of individual database roundtrips per second, rapidly causing connection pool exhaustion, write-lock contention, and latency spikes exceeding 250ms.
The Architecture: Downlink frames arriving over WebSocket or UDP are buffered into high-resolution micro-batch windows (50ms–100ms). Rather than executing N network roundtrips, the edge ingestion worker packs typed columnar arrays into a single parameterized PostgreSQL query utilizing unnest(). This cuts database write latency by >85%, eliminates lock starvation, and guarantees zero packet drops under high-rate burst.
The Challenge: Asymmetric cryptographic algorithms (like RSA or ECDSA signature verification) introduce 1ms–5ms CPU compute overhead per packet, causing unrecoverable queue backpressure on high-frequency telemetry streams.
The Architecture: We employ hardware-accelerated symmetric HMAC-SHA256 signing (via Web Crypto API at the browser/edge and Ring-3 userspace primitives in Rust via anti-cheat-sdk). Frame headers contain monotonic sequence IDs, timestamp bounds, and cryptographic MAC digests. The ingest parser computes the hash in <0.08ms per frame, rejecting replayed packets or out-of-order jitter before state propagation.
The Problem: Giving autonomous agents open-ended natural language prompts or loosely typed tools results in prompt drift, schema hallucination, and runaway tool recursion loops under complex multi-step failures.
The Architecture: We isolate agent capabilities behind strict Model Context Protocol (MCP) servers with runtime Zod/Pydantic JSON schema boundaries. Every tool call is deterministic, strictly validated, and executed with idempotent mutation tokens. If a tool fails, deterministic error metadata is fed back to the model rather than raw stack traces, allowing self-correcting reasoning loops capped by finite state machine budgets.
The Failure Mode: Third-party payment gateways (Stripe Connect, interbank ACH/Fedwire) retry webhooks asynchronously on network timeouts. In naive systems, non-idempotent webhook handlers result in duplicate payouts and unbalanced balances.
The Architecture: Multi-tenant ledger architecture engineered with strict double-entry bookkeeping in PostgreSQL. Every monetary transaction requires matching debit and credit entries committed atomically under serializable isolation. Inbound webhooks pass through cryptographic signature verification (stripe-webhook-guard) with atomic deduplication keys, ensuring idempotent processing even under aggressive retry floods.
From regulated biometric security pipelines and national ATM banking networks to low-latency Rust & TypeScript cloud infrastructure.
anti-cheat-sdk): Engineered userspace Win32 process memory inspection (K32GetModuleFileNameExW), SHA-256 binary integrity hashing, sub-ms input timing entropy detection, and HMAC-SHA256 signed event streams.unnest() bulk inserts and automated anomaly detection rules.stripe-webhook-guard).Production fluency across systems programming, full-stack TypeScript, distributed data pipelines, and agentic AI.
I build AI systems, own the full loop from engineering to customer, and move fast. If that's your team — let's talk.
I'm actively open to the right opportunity. Reach out directly — I respond fast, I'm direct, and I respect your time.