Co-founder & AI Lead of TrueAI|Trading Co-pilot and Founding Partner, AI of INC4, building at the intersection of AI, trading, and Web3 infrastructure.

_{Robotics · Embedded & field communications · Geospatial systems · AI/data reliability · Web3 infrastructure}

I work on focused reliability fixes across robotics, embedded and field communications, geospatial systems, and AI/data tooling.

• open-rmf/rmf_ros2#543 — fixed the phase key used to publish skip requests.
• pylessard/python-can-isotp#160 — handled reserved STmin values according to ISO 15765-2.
• fsspec/filesystem_spec#2102 — kept live block-cache instances usable after pickling.

• AI/data reliability and compilation: joblib/joblib#1840, apache/tvm#20141
• HTTP networking reliability: urllib3/urllib3#5186 — rewinds seekable file-like request bodies before PoolManager redirect resends.
• Geospatial I/O: geopandas/pyogrio#693
• Robotics, autonomy, and field communications: aerostack2/aerostack2#987, brian7704/OpenTAKServer#363, brian7704/OpenTAKServer#364, PointCloudLibrary/pcl#6464, PointCloudLibrary/pcl#6465, BehaviorTree/BehaviorTree.CPP#1191, isl-org/Open3D#7536

_{These items are open and under upstream review; they are not listed as merged contributions.}

• aiofiles#226 — approved exact-head async iteration and context-manager support for scandir after base RED, head GREEN, and 222 passing project tests.
• BehaviorTree.CPP#1185 — approved a trace-write clamp after an exact-base ASan oracle reproduced an out-of-bounds read on long event names.
• BehaviorTree.CPP#1186 — approved the integer-width preservation fix after checking its dispatch boundary and regression coverage.
• BehaviorTree.CPP#1192 — reproduced a root-blackboard remap crash and approved the exact-head fix after a local RED/GREEN check.
• fsspec#2106 — reproduced async deletion beyond maxdepth and approved the exact-head boundary fix after the full async target passed.
• fsspec#2103 — reproduced an archive member escaping the requested local destination and approved the exact-head containment guard after current-master integration tests.
• jsonschema#1522 — approved the exact-head fix that distinguishes no patterns from a valid empty regex after base RED, focused, full-suite, and current-main integration checks.
• redis-py#4288 — approved the exact-head async write guard after base RED, a 10-case close-race and error-translation matrix, current-main integration, and 368 green live checks.
• fsspec#2101 — proved that the async filecache range path still redownloaded the full object because successful downloads never entered the metadata cache.
• AnyIO#1294 — found that the first OS-refused datagram still completed successfully while the new backpressure only blocked the following send.
• ImageIO#1203 — caught a broad exception handler that hid invariant and plugin failures from the project's image fuzzer.
• Shapely#2465 — found a pickle/deepcopy precision-restoration path that erased pointwise-collapsed line geometry instead of preserving its coordinates.
• PyVista#8957 — found that malformed negative legacy cell counts could trap the new diagnostic parser in a non-progress loop instead of raising.
• pydicom#2337 — found that an unknown-VR error translation also mislabeled documented extension-hook failures as malformed DICOM input.
• python-can#2089 — caught an unrelated workflow change that replaced two immutable Coveralls Action commit pins with a mutable release tag.
• pyserial#868 — found that narrowed RFC 2217 exception handling leaked an allocated socket and reader thread when an interrupt aborted port reconfiguration.
• msgspec#1164 — found that a metaclass-selection fix bypassed custom construction hooks and their documented struct configuration.
• nats.py#1007 — found a supported token-only credential form left exposed by a log-redaction patch and requested a focused regression.
• OpenTAKServer#362 — caught a syntax failure and locale mappings to unavailable catalogs before merge.
• pyogrio#691, cantools#830, and OpenTAKServer#349 — reproduced merge-blocking error-handling and shared-state regressions and submitted concrete CHANGES_REQUESTED reviews.

Building AI-native trading infrastructure.

• Native AI agents for on-chain trading
• Focus on execution logic, risk systems, and transparency
• Turning AI from analysis tools into decision + action systems

Trading is the entry point.
The long-term goal is trustable autonomous systems for financial markets.

Building and using foundational infrastructure for AI-driven on-chain systems.

• Reliable infra for AI agents operating on-chain
• Performance, availability, and composability
• Infrastructure designed for real-time autonomous execution

• Build products that make complex systems usable
• Turn intelligence into real, on-chain outcomes
• Help AI and Web3 create real value, not just narratives

• Twitter
• LinkedIn