Shipping production LLM
and multi-agent systems.

Promoted from Data Engineer / Full-Stack Developer (2022)

• Built an internal Model Context Protocol (MCP) server unifying MySQL, AWS S3, Google Drive, and Google Sheets behind a hybrid RAG + MCP interface. Analysts ask questions in natural language and get immediate SQL execution, default visualizations, and a custom Power BI–style dashboard builder — replacing a workflow that previously required a developer to write queries and build reports manually.
• Shipped the MCP platform as a Django application with a custom UI; stack: FastMCP, FastAPI, vector DB for hybrid retrieval.
• Designed and deployed a multi-agent SEO content system — ~8 autonomous CrewAI agents covering research, analysis, scraping, and writing — running the full content pipeline end-to-end (CrewAI, FastAPI, Docker, Pydantic).
• Built three production RAG systems: (1) a cross-source knowledge RAG spanning Google Drive, Slack, internal docs, and external conference and article content, with full source attribution for every answer; (2) a competitor content intelligence RAG for angle and gap analysis against competitor articles and site structures; (3) a historical content performance recommender indexing past articles against their metrics so writers can surface what worked for a given topic with evidence.
• Established evaluation and observability for the RAG and agent systems using LangFuse and LangSmith — tracking groundedness, answer quality, latency, and cost per query across OpenAI, Anthropic, and Gemini.
• Own architecture decisions, model selection, prompt design, guardrails, and cost/latency tuning.
• Partner with product, marketing, and data teams to turn vague problems into shipped AI features.

• Build and ship a production AI application that generates compliance reports — ingests client materials, evaluates them against regulatory requirements, and returns a structured report flagging compliant and non-compliant areas with evidence.
• Designed the end-to-end pipeline from ingestion and retrieval through inference, evaluation, and monitoring — with a focus on reliability, low-latency inference, and auditable outputs.
• Implemented guardrails and evaluation harnesses for groundedness, citation accuracy, and hallucination rate, so every report claim is traceable to the source regulation.
• Optimize for cost and performance through batching, caching, vector search tuning, and async processing; established LLM observability across requests, tokens, and latency.
• Own architecture decisions, production debugging, and post-mortems across the stack.

• Built and shipped a multi-agent AI customer support system serving ~500,000 users in production. Roughly 10 autonomous agents handle refunds, how-to support, ticket triage, bug reports, and credit-related queries.
• Designed across multiple repositories (agent service, support backend, infrastructure) using FastAPI, Docker, and AWS ECS.
• Integrated multiple model providers — AWS Bedrock, OpenAI, Anthropic, Gemini — with routing logic balancing cost and latency per agent role.
• Owned the full pipeline: ingestion, retrieval, inference, tool use, guardrails, and LLM observability (tracing, latency, cost, quality metrics).
• Built evaluation harnesses over real production conversation data to catch regressions in agent behavior; iterated prompts and routing with product and backend teams.

• Architected scalable ETL pipelines with Pydantic, dbt, and Airflow — 30% faster processing and 20% operational efficiency gain for enterprise clients.
• Tuned Postgres and DuckDB analytical workloads through query optimization and indexing strategies, accelerating decision-making by 20%.
• Enhanced Snowflake warehouse performance and reduced query costs on high-volume analytics.
• Built 10+ reproducible containerized development environments (Docker, Poetry), cutting onboarding time by 40%.
• Delivered interactive Power BI and Plotly Dash dashboards with cross-filtering, callbacks, and drill-downs.
• Enforced testing (pytest), type safety (mypy/pyright), and code quality (ruff/black) across production codebases.

Promoted to Senior AI Engineer in Nov 2024

• Shipped 14 production Django + Plotly Dash applications for 50+ internal users, eliminating ~80% of manual reporting workflows.
• Designed LogAna, an automated log analysis platform processing 10M+ daily log entries — cut manual analysis from 8 hours to 30 minutes (400% improvement in processing speed).
• Engineered high-throughput data pipelines (pandas, Polars) processing millions of daily records with 99.9% uptime.
• Built ETL workflows across REST APIs, AWS S3, PostgreSQL, and DuckDB with sub-second query performance.
• Implemented asynchronous web scraping services with proxy rotation, rate limiting, and robust error handling.
• Managed AWS-hosted PostgreSQL with Redis caching and strategic indexing for fast, reliable data access.

Top Rated Upwork Freelancer · 100% Job Success Score

• Architect and deploy production LLM applications — LangChain, LlamaIndex, RAG with Pinecone and PGVector — delivering document Q&A and semantic search for global clients.
• Build custom AI automation workflows integrating GPT-4, Claude, and open-source models; clients report ~60% operational efficiency gains.
• Design end-to-end ETL pipelines for lead generation and marketing analytics — 60% accuracy improvement and 70% processing time reduction.
• Founded knowtheprice.com.au — full-stack real-time price comparison platform (Next.js, Supabase, Vercel).