Goblin Reactor
Systems software consulting
Systems performance consulting
Goblin Reactor diagnoses, designs, builds, tests, and rolls out difficult performance and infrastructure systems—with measured results in memory, latency, throughput, reliability, and cost.
Send a short technical intake. Qualified problems receive a complimentary 20-minute fit call.
C++, Python, SIMD, market data, caching, search, matching, and performance-critical backend infrastructure.
The Delivery Gap
Some teams lack the time or specialist capacity to build the system at all. Others can generate implementation quickly but remain stuck on architecture, integration, correctness, migration, performance, and production rollout.
Goblin Reactor takes ownership of the hard technical path: establish what is true, design the system, build or direct the implementation, test the ugly cases, manage the rollout, and measure the result.
What Goblin Reactor Fixes
Goblin Reactor works in the uncomfortable layer where software meets the machine—and where promising projects become expensive operational problems. That includes architecture, data layout, cache behavior, vectorization, memory pressure, serialization, storage tiers, migration, rollout, and the difference between code that works in isolation and a system that keeps working under load.
Services
01
Diagnose the real constraint, establish a production path, and deliver a working result for systems that are slow, expensive, unreliable, stalled, or too technically awkward for ordinary consulting.
02
Replace pointer-heavy, general-purpose infrastructure with compact workload-specific designs that keep behavior while wasting less RAM.
03
Scalar reference validation, vectorized kernels, runtime dispatch, benchmark harnesses, and performance regression tests.
04
Infrastructure for tick data, replay, parsers, latency-aware research, real-time indicators, and execution-adjacent tooling.
About
Goblin Reactor is Adam DePrince, a systems and performance specialist who has been building production software since 1989, when he wrote Pascal for a pharmaceutical manufacturer while still in high school. Since then his work has spanned startups, big tech, and the stranger corners of high-performance and market-data infrastructure: quantitative and market-data systems at Quantlab and Massive.com, financial data at scale at S&P Global, and large-scale engineering at Google and Amazon. The open-source projects below are his—and they are the proof behind the consulting offer. Based in Vermont; available for remote work.
Proof From Open Source
Redis-compatible sorted sets and hashes built on compact Swiss-table indexes and packed arenas. In parity benchmarks, Goblin Core is the leanest engine tested and beats every tested Redis-compatible server across measured sorted-set operations, while also leading the hash memory comparison.
A large-object cache and HTTP object server that uses RAM, SSD, and hard drives together to replace expensive RAM with SSD and expensive SSD capacity with hard-drive capacity. For 256 KiB-8 MiB objects, Goblin Store uses 7.4x less RAM at memcache latency, which means substantially lower costs.
SIMD-accelerated sequence alignment and fuzzy matching with Unicode/CJK support and runtime dispatch across multiple CPU families.
Massive.com flatfile parsing and market microstructure simulation, iterating flatfiles faster than Python gzip and csv.
C++ and Python Kalman filters running dramatically faster than OpenCV on small fixed-size filters used in streaming workloads.
Architecture-specific acceleration work enabling NumPy's 256-bit Loongson SIMD path and extracting speed from an unfamiliar stack.
More proof-of-work lives at adamdeprince.com.
Delivery
01
Goals, stakeholders, constraints, profiles, traces, benchmarks, failure evidence, and business impact.
02
Architecture, ownership, milestones, interfaces, risks, migration, and acceptance criteria.
03
Direct implementation or technical direction of AI and human contributors.
04
Correctness, failure recovery, concurrency, compatibility, load, and performance regression testing.
05
Production migration, operational validation, and comparison against the agreed result.
Engagements may cover one stage or the entire path.
Technical Intake
A few concrete details are enough. I am looking for the shape of the system, the evidence already available, the business cost, and whether the people who can act on the answer are involved.
I review serious submissions personally. When the problem and evidence look like a fit, I may reply with an invitation to a complimentary 20-minute call. The scheduling link is sent only after intake review.
Engagements
If Goblin Reactor does not deliver the documented profile, headroom analysis, and prioritized recommendation, the audit fee is refunded.
After reviewing your intake, I may invite you to a short call to determine whether the problem, evidence, budget, and working relationship are a fit.
This call is for qualification, not debugging or architecture work. It is not a free consulting engagement.
Invitations are sent selectively after intake review. If Adam invited you, schedule the 20-minute fit call.
A focused 90-minute technical session with advance review of the available evidence, discussion of the system's constraints and likely failure modes, and a written proposed investigation plan.
A focused one-week investigation that leaves the client with a documented performance profile, a headroom analysis, and a prioritized recommendation, even when that recommendation is "leave it alone."
Scheduled delivery work for a defined systems problem, from implementation through testing, migration, rollout, and measured production validation.
Architecture, data structures, storage layout, and performance-risk review before a design becomes expensive to unwind.
Engagement Structure
Engagements are scoped around the problem and expected business value.
Work may begin with a paid technical scoping session or focused performance audit, followed by a defined implementation engagement when the evidence supports it. Smaller diagnostic or prototype engagements are available when they are the right first step.
Goblin Reactor does not provide open-ended staff augmentation, emergency on-call support, or commodity hourly development.
Good Fits
Client code, data, credentials, and other confidential material are handled under written confidentiality and security terms.
The development environment—including any AI tools, data-retention settings, and dedicated-hardware requirements—is agreed before work begins.
I do not submit client material to services that use it for model training. Where client policy requires, work can be performed on dedicated hardware without external model access.
Contact
Start with the short technical intake. Qualified problems may receive a complimentary 20-minute fit call. Available for remote consulting from Vermont.