Shafir Khajo

Built a real-time environmental evidence system with immutable revisions, multidimensional validation, deterministic policy rules, provenance tracing, adaptive monitoring, offline field capture, H3 stream-reach modeling, FHIR interoperability, transactional workflows, and bounded AI for extraction, explanation, and media triage.

Built and integrated the backend for a live synthetic exchange with autonomous traders, price-time matching, risk controls, spoofing surveillance, evidence-backed citations, and adversarial evaluation. Detected the baseline spoofer across 171/171 held-out market paths while cutting median order latency from 15.3 ms to 1.3 ms.

Built a fully local experimentation system that generated, compiled, and parallel-tested autonomous bot variants across controlled maps and seeds. Candidates passed progressively harder evaluation stages before facing the incumbent champion, while a local Qwen3.5-9B model proposed new strategies from prior results.

Built an open-source TypeScript decision engine for vehicle and marine ownership costs, combining AI-assisted listing extraction with deterministic financial models for depreciation, financing, operating costs, uncertainty, and scenario analysis. Cars and boats use separate engines with shared schemas, versioning, and reporting contracts.

Led architecture and integration for a winning counter-UAS sensor-fusion platform combining heterogeneous detections into confidence-aware threat assessment and operator decision support. Designed the system to surface evidence quality, uncertainty, prioritization, and situational context for airport and critical-infrastructure defense.