Selected public AI systems and engineering work
The portfolio is organized by problem and system type. Public repositories are evidence of implementation scope; they are not presented as client engagements or measured business outcomes.
Project categories
AI Agents · AI Automation · RAG / Knowledge · Document AI · Business Systems · Developer Tools · Analytics · SaaS / Internal Tools. Each flagship card names the problem, architecture, role, stack, evidence and related service.
MCP Composer — AI Agents
Problem: agent toolchains become opaque. Solution: a reviewable MCP tool registry, permission model, workflow builder, dry-run simulator, approvals and audit history. Role: system framing, architecture and implementation. Stack: FastAPI, Python, SQLite and a responsive web UI. Outcome: public implementation evidence; no client metric claimed.
Model Router — AI Systems
Problem: requests need different model capabilities, latency and cost policies. Solution: an explicit provider-routing layer with configurable selection logic and observable decisions. Role: architecture and implementation. Stack: Python, FastAPI and model-provider adapters. Outcome: inspectable routing system; no production savings claimed.
Universal Document Parser — Document AI
Problem: heterogeneous documents do not share one reliable schema. Solution: ingestion, classification, schema mapping, extraction, validation and structured JSON output. Role: architecture and implementation. Stack: Python, FastAPI and document-processing components. Outcome: demonstrable workflow; accuracy requires a representative benchmark.
Invoice AI — Business Automation
Problem: invoice formats vary while accounting needs structured fields. Solution: extraction, normalization, validation and a human-review boundary. Role: product framing and implementation. Stack: Python, FastAPI and OCR/model integrations. Outcome: public invoice-to-fields workflow; no unsupported accuracy claim.
Supplier Email Parser — Business Systems
Problem: procurement data is buried in messages and attachments. Solution: parse approved threads into structured records for operator review. Role: workflow design and implementation. Stack: Python, FastAPI and extraction services. Outcome: public implementation artifact; no client impact claim.
AI Test Case Generator — Developer Tools
Problem: requirements must become consistent, reviewable test scenarios. Solution: structured candidate generation with engineering review kept authoritative. Role: architecture and implementation. Stack: Python, FastAPI and LLM integration. Outcome: inspectable developer workflow; generated cases still require review.
API Docs Generator — Developer Tools
Problem: API behavior and examples drift from manually maintained documentation. Solution: derive reviewable documentation from definitions, schemas and examples. Role: system framing and implementation. Stack: Python, FastAPI and OpenAPI processing. Outcome: publishable documentation artifacts with human verification.
Runbook Generator — Operational AI
Problem: incident knowledge is fragmented under pressure. Solution: turn approved service context into draft runbooks with owners, checks and escalation. Role: workflow and product implementation. Stack: Python, FastAPI and structured generation. Outcome: reviewable operational artifact; not a substitute for tested incident procedures.
More Projects / Experiments
The complete public GitHub profile includes additional bounded experiments across automation, analytics, document processing, safety and internal tools. Repository status and limitations remain the source of truth.
Contact →
Keep scrolling to enter the next page and see every card inside it.