BuildingML Systems

SchemaPulse

OpenAI-compatible inference layer for schema-valid structured outputs with streaming tool-call parsing.

🎯The Problem

Structured outputs often fail in real products: streaming partial JSON breaks parsers, tool-call arguments drift from schema, and strict guided decoding increases latency/throughput cost.

💡The Solution

Schema-first outputs using Pydantic models. Streaming incremental parsing that validates progressively. Adaptive guided decoding that tightens constraints only when drift is detected.

✨Key Highlights

  • Streaming incremental tool-call parsing
  • Adaptive guided decoding for schema validity
  • Live UI for validation status

Other Projects

HyperSentry

An agentic security copilot that uses policy checks + retrieval to generate remediation PRs and accelerate incident analysis.

EdgeJury

Multi-LLM council with cross-review for truthful QA on serverless edge inference.

TariffWhisperer

End-to-end RAG system over CBP rulings for explainable HTS code recommendations.