I build production-grade AI systems that combine rules, workflows, and LLM intelligence—focused on correctness, scale, and real business outcomes. But My Primary Skill is Figuring it out and making it work.
Strong inclination toward:
“Even before AI, I was interested in how components interact, fail, and scale.”
Built Critical Path Methodology (CPM) engine handling:
Impact: Optimized workflow duration by 40% and storage usage by 30% (deduplication).
Where my systems brain formed: modeling time, constraints, and graphs.
Building Hybrid Rule + LLM validation engines & deterministic workflows.
Focus: Accuracy (98%), Performance (<10s), Auditability.
CPM engines, Rule-based systems, Transition graphs, Parallel execution models.
Layout analysis, Table detection, Section extraction, Deduplication & post-processing.
RAG pipelines, Hybrid validation engines, Prompt ensembles, Semantic extraction.
FastAPI microservices, Async & parallel processing, Secure metering & monetization.
Manual regulatory validation is slow, ambiguous, and error-prone.
30+ deterministic rule checks combined with LLM-based semantic validation and ensemble logic to resolve ambiguity.
Input Docs → Rule Engine → Semantic Validator (LLM) → Decision Aggregator → Audit Logs
PDFs are messy. RAG pipelines fail without high-quality preprocessing.
Intelligent chunking, TOC extraction, and vectorization pipeline with parallel processing via Celery.
Retrieval quality matters more than the LLM.
Standard workflow engines fail when dependencies become complex.
Graph-based CPM engine supporting parallel activities, routes, custom dummy activities, and forward calculation logic.
Logic & Mathematics applied to real-world time modeling.
Pure LLMs are non-deterministic. In compliance and high-stakes fields, 98% isn't enough without a safety net. My approach pairs strict deterministic pre-checks with LLM semantic reasoning only where necessary.
Hardcoding workflow steps creates technical debt. I build engines where the workflow itself is a graph data structure, allowing dynamic routing and easier maintenance.