02
ENGINEERING WORK
FEATURED PROJECT
POLICYGUARD AI V2.0
<50ms
Interception Latency
0-Trust
Architecture
Gemini 3
Powered
ENGINEERING DECISIONS
- Cascading Model Architecture: Fallback across confirmed-available Gemini 3.1 models for maximum uptime.
- Zero-Trust Interceptor Proxy: Every prompt screened by Gemini 2.5 Flash-Lite in under 50ms.
- GraphRAG Policy Cognition: NetworkX-powered knowledge graph detecting cross-document conflicts.
- Autonomous PR Synthesis: Auto-creates branches with guardrail-patched code via LangGraph loops.
FEATURED PROJECT
LEXINEL
◆ RUNNER-UP · HACKFEST 2.0 · NATIONAL
10K+
Records Real-Time
−60%
Query Latency
RAG
+ LangGraph
ENGINEERING DECISIONS
- LangGraph agent orchestration with hybrid AI + rule-based compliance reasoning — deterministic where policies are explicit, LLM where ambiguous.
- RAG pipeline with semantic search over policy docs via vector retrieval; every output grounded against retrieved context — hallucination surface slashed.
- Redis caching on hot compliance query paths drove −60% latency at 10k+ transaction scale — no schema changes, pure caching strategy.
- FastAPI async architecture: typed contracts, connection pooling, structured error propagation for concurrent audit workloads at scale.
ML SYSTEM
SLA-GUARD AI
Logistic regression on engineered features — SLA burn rate, latency trend velocity, error rate acceleration — for early breach detection. Explainability module + real-time FastAPI alerting.
ORCHESTRATION ENGINE
MERIDIAN
Event-sourced workflow orchestration with rollback recovery — agent failures recoverable, full audit trail by design. AWS SQS distributed task queue, Cognito auth across multi-service architecture.