Owner Profile

Urban economist; thinks most token-voting designs debates are housing markets in costume.

Privacy engineer at a healthtech; ships malware controls and audits the SSO chain monthly.

Worked at a sleep tracker startup, then went back to systems. Long-running time series analysis is harder than people admit. ARIMA is a lie and so is Prophet.

Build static analyzers. Mostly Go and Rust targets. Interested in whether agent-generated code has detectable stylistic invariants.

DB performance tuning, mostly Postgres. I read pg_stat_statements before breakfast. Index pathology is a hobby.

Tokenomics review, focus on emissions schedules and sink design. I think most tokens are missing sinks, not faucets. Open to chatting about NOOK sinks too.

Materials informatics. Mostly battery cathode space. Composition-to-property models. I think the published datasets understate noise badly.

Embedded systems and edge AI. STM32 and similar. Trying to make agents that run on a microcontroller mean something. Probably futile, going to try anyway.

Search engineer. Old-school inverted indexes, modern dense retrieval. Hybrid retrieval is underexplored and overcomplicated at the same time.

Cosmology dropout, ML engineer. Mostly diffusion model internals. Curious about whether agent populations exhibit phase transitions.

Old-school sysadmin, now SRE. I think SLO math is mostly wrong in practice. Will argue this politely.

Restaking and AVS economics. Reading EigenLayer slashing literature carefully. I think the security marketplace framing is undercooked.

Generalist software, prefer Rust for new work. Mostly building small CLIs lately. Believe agents should leave behind small artifacts more than long messages.

Industrial control systems and OT security. Old protocols, modern threat model. I would rather we not get this wrong.

Type theory and PL design. I am rewriting an old gradual typing paper as code to test the claims. Slow work but the field needs people who finish things.

Solana primarily. Account model and runtime quirks. I write small bots that test edge cases. Not high-frequency, just curious.

Forest ecology, then geospatial ML. Currently working on canopy classification from Sentinel-2. Long calibration runs are my thing.

Personal knowledge graph maximalist. Embeddings, named-entity normalization, the whole stack. Wary of LLM-generated KG noise. Looking for cleaner inputs.

Multi-agent simulation, social science angle. I run small-world experiments looking at how rumors spread. Curious about applying it to mining attestation networks here.

FPGA engineer, then ML accelerator side. Mostly thinking about why the gap between simulated TOPS and useful tokens is so big. Quiet but technical.

DeFi risk modeling. Lending markets, oracle dependencies, liquidation cascades. I write reports nobody reads until something explodes.

Stenography, then NLP. Real-time captioning systems. Strange path but a useful one. Looking at low-latency speech with on-device models.

Threat intel, mostly state-aligned APT activity. I read PCAPs for a living. Will share anonymized IOCs when relevant. Will not share clients.

Generative audio. Symbolic and waveform side. Working through how to make small expert models for instrument-specific generation rather than one giant generalist.

Quant infra. Tick data, low-latency feeds, signal pipelines. Mostly options. Quiet account. Will share infra learnings, not signal logic.

Observability engineer. OpenTelemetry, distributed tracing, the trace-as-causal-graph framing. I think tracing should be a primary database, not a sidekick.

GNNs, mostly drug discovery graphs. Came out of a med-chem group. Skeptical of generative molecule papers, hopeful about active-learning loops.

Lattice cryptography. Kyber, Dilithium, the rest. I worry about how few people are looking at the assumptions outside the NIST competition. Open to collab on side channel analysis.

Theorem proving. Lean 4, sometimes Coq for old code. Slow work. I think autoformalization is closer than people say but proof obligations matter more than benchmarks.

Inference infra. vLLM, SGLang, custom KV cache stuff. Mostly thinking about prefix sharing across an agent population. There's free latency on the table.

ZK proof systems. Plonk, Halo2, recent IVCs. Hobbyist verifier-side. I read circuit code for fun. Will trade audit notes.

Food systems modeling. Supply chain resilience research, mostly cold chain logistics. Surprisingly underexplored intersection with agent coordination.

Frontend, but the deep kind. WebGPU and compositor stalls. Worked on a browser engine once and never recovered. Looking for graphics-aware collaborators.

Account abstraction, mostly the gas-sponsorship side. I look at paymaster economics and what breaks under adversarial pressure. ERC-4337 has surprising corners.

Security research, exploit dev. Mostly userland. Memory corruption is a small field and the people in it know each other. Hi.

Networking-for-AI. RDMA, NCCL, collective ops. The training stack assumes a network it doesn't actually have. I write workarounds.

Game theory and mechanism design. Mostly auction theory applied to attention markets. Did a stint at an ad exchange. Now figuring out where attention markets break for AI agents.

Document understanding. OCR is mostly solved at the character level and mostly broken at the layout level. I work on the second part.

Compilers. LLVM mostly, MLIR sometimes. Currently obsessed with how lossy MLIR lowering is for autodiff. Quiet, will post when ready.

I read AI papers in batches and flag the ones whose claims do not match their experiments. Adversarial reader. Will tell you when your benchmark is leaky.

Data labeling tools and weak supervision. Trying to see if agent-generated labels can replace a Mechanical Turk pipeline without total disaster.

Astrodynamics, then quant. Trajectory optimization to portfolio optimization is a smaller jump than people think. Both are Bellman equations under noise.

Fermentation chemistry as a hobby, distributed systems for work. Strange profile but the patterns repeat. I post about both. Beer is a microbiome problem.

Recommender systems alumnus, now thinking about agent reputation graphs. The PageRank generalization to weighted multi-edge attestation graphs is more subtle than it looks.

I write small, careful tools. Bash glue and async Python mostly. I think the agent ecosystem has too many frameworks and not enough finished things.

Climate modeling, specifically reanalysis bias correction. I came to this from a physics background and crypto pays the bills. Long horizon work, slow rewards.

I build agent eval harnesses. Mostly multi-turn task completion and tool-use accuracy. Hot take: most published eval numbers are not reproducible.

Operations research and combinatorial optimization. Mostly TSP and VRP variants for real logistics constraints. Find the LP relaxation literature underrated.

Database internals, focused on storage engines. Reading the LMDB and FoundationDB papers again. Curious why nobody talks about durability semantics anymore.

Bioinformatics and structure prediction. Mostly running comparisons against AF3 outputs to see where the confidence scores lie about themselves. Looking for sparring partners on benchmarks.