I build high-throughput data ingestion pipelines, deterministic verification runtimes, and policy-as-code guardrails for autonomous machine workflows.
Most security tools rely on slow, probabilistic LLM-based evals that add massive latency and remain vulnerable to prompt injections. My work focuses on moving verification to the transport and execution layers:
- Agnostic Serialization: Intercepting raw JSON-RPC 2.0 and HTTP streams (
stdio/SSE) flowing between AI hosts and execution environments. - Structural Parameter Profiling: Flattening unstructured text, files, and payload tokens into rigid schemas in real-time.
- Deterministic Policy Compilation: Compiling local Open Policy Agent (OPA/Rego) rules to enforce strict security boundaries in under 3–5ms with zero hallucination.
- Cryptographic Provenance: Hashing state transitions via multi-key SHA256 ledgers for immutable forensic compliance and auditing.
- Data Systems & Pipelines: High-volume normalization, distributed task orchestration (Celery, Redis), and indexing architectures.
- Runtime Sandboxing: Enforcing execution-layer contracts, credential isolation, and parameter firewalls.
- Computational Efficiency: Optimizing low-latency streaming evaluation engines and algorithmic throughput.
- Languages: Python, Java, SQL, TypeScript, Go (exploring)
- Infrastructure & Backends: FastAPI, Postgres, Redis, Docker, AsyncIO, Linux, Open Policy Agent (OPA)
- LinkedIn: linkedin.com/in/bhuwanbhandari99
- Portfolio: bhandaribhuwan.vercel.app
- LeetCode: leetcode.com/u/beebeeVB/