Applied AI Engineer - AI Coding Agents and LLM Systems

AI Agents and LLM Applications

Build and ship LLM agents end to end: tool calling, multi-agent orchestration (LangGraph, LangChain), structured outputs, MCP, and provider-agnostic layers over OpenAI, Anthropic and Google models. Built PocketPal AI (open source, 4M+ downloads, 300k MAU), whose tool-calling agent, web search and document chat run fully on the phone.

RAG and Enterprise AI

Agentic RAG for regulated industries. At H2O.ai, replaced a top-10 global bank's RAG system (~70% on its own benchmark) with an agentic pipeline: VLM-based document parsing, structured storage and query routing; reached 95% on the bank's validation set. Also built policy-aware banking agents and LLM-based document validation.

LLM Evaluation and Benchmarking

Evals that drive product decisions: rubric-based LLM-as-judge, failure taxonomies, tracing (LangSmith), replay harnesses, holdout testing. ferret-bench raised web-search agent correctness from 0.79 to 0.92, and that configuration ships in PocketPal; dogear-bench cut hallucinated answers in document chat by 9 points.

AI Coding Agents and Agentic Engineering

Software delivery through coordinated coding-agent teams on Claude Code, Codex, opencode and pi: GitHub issue to reviewed draft PR, with worktree isolation, permission policies, hooks and human-gated production changes. Open-sourced pocketpal-dev-team and agistry (agent coordination, hand-off messaging over MCP, leases on shared test devices and GPUs).

On-Device AI and Fine-Tuning

LLMs on mobile hardware with llama.cpp, ExecuTorch, LiteRT-LM and quantization; cross-engine benchmarking on CPU/GPU/NPU (xebench, public leaderboard at pocketpal.dev). Contract at Meta: fine-tuned and deployed small language models on device for NLP use cases.

Software Engineering and Cloud

Python, TypeScript, React Native, React, Go. Backend services on GCP (Cloud Run), AWS, Supabase and Kubernetes; CI/CD and app-store releases.

Machine Learning and Data Science

10+ years with XGBoost, scikit-learn, TensorFlow/Keras, Spark MLlib, H2O and DataRobot: predictive modelling, churn, fraud detection, demand forecasting and predictive maintenance in banking, insurance, manufacturing, automotive and logistics. Technical lead on 10+ enterprise POCs at DataRobot; built in-house data science platforms at Volkswagen (AWS, Cloudera, Open Telekom Cloud).