The full list

All projects.

Everything, in one place: systems, ML, agents, studies, and hackathon builds. Every number traces to a committed benchmark, test run, or published result, and seventeen of these run in your browser on the repository’s own code.

strata LSM-tree storage engine in C++20: WAL, MVCC, group commit, crash-verified durability
11,149 SIGKILLs · 2.22M acked writes, 0 lost · 50× RocksDB on sync commits run it ↗ taut Reliable-UDP transport + SWIM membership for lossy links, losing axis published
p99 ~6× below kernel TCP at 5% loss, ~12× at 20% run it ↗ tautq Coordinator-less distributed webhook delivery on taut: WAL replication, epoch-fenced leases
800 jobs/s · p99 107ms · 100% completion at 0-20% loss run it ↗ Tollgate Multi-tenant API gateway for shared LLM keys: atomic Redis limits, OTel, Kubernetes
2,000 req/s, 0 errors, p99 46ms · global limit held across 3 replicas run it ↗ Model Cascade Verify-then-escalate routing over 100 frozen EvalPlus tasks, published by GMI Cloud
94/100 vs best single model's 90 · 47.1% cheaper post ↗

Mint Lead-intelligence platform at GMI Cloud: identity resolution, LLM ICP tiering, 16-tool agent
18K LOC · 45 endpoints · 251 tests · 51 qualified leads internal

Winnow Voice-first LLM prompt compression on two orthogonal axes (token-space + model-space)
~60× KV reduction · F1 0.437 vs 0.376 fp16 on LongBench repo ↗ Pylon Counter-UAS RF sensor mesh: anomaly detection, dual SDR backends, 3D operator dashboard
catches novel emitters signature baselines miss · mesh converges ~2s repo ↗ Vigil Autonomous AI on-call engineer: zero standing credentials, single-use scope-bound grants
5 services · 60s-TTL grants · full audit trail repo ↗ LOCUS Voice-driven search-and-rescue command center: Monte Carlo heatmap, LLM tip reweighting
150 walks/hypothesis × 4 ISRID profiles · 1.5s tick, zero LLM in-loop repo ↗ AgentBench Eval framework for LangGraph agents: cost-adjusted accuracy, Ray Tune HPO, CLI + leaderboard
0.964 acc · n=55 · $0.042 total spend run it ↗ Checkit Health 7-stage misinformation-detection pipeline over 4 live social APIs, cost-tiered LLM path
64 tests · 15 req/min free tier held repo ↗ Argus Agentic training monitor: SPC detection across 4 failure modes, bounded self-correction
49% of anomalies auto-resolved repo ↗ SitRep Team project-management app: led 11 engineers through a 10-week Agile SDLC on Cloudflare
11 engineers · 5 sprints · shipped live live ↗ ChainSense Ethereum behavioral analytics: 16-feature wallet matrix, 9 archetypes, anomaly flagging
403K txns · 59K wallets · 591 anomalies repo ↗ RepoMap VS Code extension mapping any repo as an interactive dependency graph with AI explanations
6 languages parsed repo ↗ AgriShield Wildfire risk + firebreak optimizer: ELMFIRE ensembles score 24 candidate layouts
92 tests · mypy-strict clean repo ↗ UrbanSound8K CNN-free audio classification: 487-dim engineered features, wavelet scattering, leak-free CV
72.9% top-1 · 0.51ms/clip inference repo ↗ Helmet Identification ResNet-50 pipeline detecting riders and classifying helmet use; published via Lumiere
94% acc · 0.92 mAP · <50ms inference · top 10% of ~200 papers paper ↗ tickvault Full-depth crypto order book archive in Rust: six venues to Parquet, reconstruction to any instant, coverage published rather than interpolated
every gap published, not filled run it ↗ decode-anatomy Where an LLM decode step actually goes on an RTX A6000: kernel time, gaps, and launch overhead separated
38.3% of a batch-1 step is not kernel work, and batching removes it run it ↗ ddp-crossover Where a second GPU stops paying on 2x T4 over PCIe, measured across model sizes
speedup crosses one going up, the opposite of the usual advice run it ↗ noisy-rag-bench What OCR noise does to retrieval versus to answers, on real scanned pages
at 5% error recall is unchanged and half the answers are already gone run it ↗ open-refusal-steering Open reproduction of feature-level activation steering, with a coherence check attached
at 100% response rate every behavioural check passes and the output is token soup run it ↗ memharness What a read-once-answer-forever memory system actually costs, priced per turn
the one read cost 12.7x all twenty questions combined run it ↗ actiongate A confidence gate and tamper-evident audit trail between a meeting transcript and a CRM write
gating at 0.65 turns 18 wrong writes into 2, keeping 19 of 22 genuine ones run it ↗ icp-benchmark 120 real YC companies hand-labeled on three axes, with a scoring harness for lead qualification
enrich-everything is 25% precision; a rules filter is 93.8% and misses half run it ↗ promptgate What happens when a published production prompt states its contract four times and contradicts itself once
the parser reads a reject as silence and the gate treats silence as approval run it ↗ pms-race Sixteen computer-use agents racing for one hotel room, with and without a transaction
the legacy path books it 93 times; the obvious fix does nothing for lost updates run it ↗ flightrisk Deposit-flight risk across every US bank, 2015 to 2026
the risk is U-shaped: a large inflow predicts a drawdown about as well as an outflow run it ↗ apparatus-comps A comparable-sales estimator for fire apparatus, back-tested against real sales
half the estimates are off by more than 36% run it ↗

Project Liftoff Incubator management platform: 6-tier RBAC via Supabase row-level security, 13 tables
100+ users onboarded private