ML / AI

Design, run, and score prompt evaluations with variance-aware benchmarks and regression tracking.

Automated multi-source research: gather, dedupe, cite, and synthesize web + paper sources into a single brief.

Build production-grade Claude API apps with prompt caching, tool use, streaming, and model migration baked in.

Build camera features the modern way — Preview, ImageCapture, VideoCapture, ML Kit overlays.

HealthKit read/write — steps, heart rate, workouts, sleep, background delivery, permissions flow.

Build LangChain agents — chains, tools, memory, RAG, LangGraph state machines, eval framework.

Production RAG — chunking, embeddings, hybrid search, reranking, caching, eval.

Migrate Claude API apps to Fable 5 (claude-fable-5) — model IDs, breaking changes, eval-gated rollout.

Prompting patterns tuned for Fable 5 — instruction style, thinking budgets, tool-use, long-context recall.

Build autonomous agents on Fable 5 — Agent SDK wiring, memory, sub-agents, MCP, cost guardrails.

Pick the right Claude model per task — Fable 5 vs Opus 4.8 vs Sonnet 4.6 vs Haiku 4.5, with cost math.

Deep multi-file code review powered by Fable 5's long-context reasoning — whole-PR understanding.

Google Gemini image API (Nano Banana / Nano Banana Pro) — generate, edit, refine, text-accurate infographics.

Build production agents on the Claude Agent SDK — agent loop, custom tools, subagents, hooks, sessions, MCP.

Generate video via Kling 3.0, Veo 3.1, and Runway Gen-4.5 APIs — model routing, image-to-video, async polling.

Drive Google's Antigravity ADE — Manager View parallel agents, verifiable artifacts, isolated-workspace review.

On-device LLM in Swift — LanguageModelSession, @Generable guided generation, tool calling, streaming, fallback.

Drive fal.ai's queue API, WebSocket streaming, webhooks, and multi-model chains across FLUX, Kling, Veo and more.

Production LLM inference on Kubernetes with vLLM + KServe — InferenceService, GPU scheduling, scale-to-zero, canary.

Engineer agent context: token budgets, dynamic selection, compression, memory tiering, and drift monitoring for long runs.

GitHub Spec Kit spec-driven development — /specify, /plan, /tasks, /implement, constitution.md, EARS specs.

Build x402 HTTP-402 paywalls and let agents pay APIs in USDC on Base and Solana via Coinbase's x402.

Persistent agent memory with Mem0 — fact extraction, add()/search() retrieval, per-user/session scoping in the LLM loop.

Launch and tune SGLang serving — RadixAttention prefix caching, structured generation, tensor parallelism, vLLM benchmarks.

Drive browsers with Stagehand's act/extract/observe — Zod-typed extraction, resilient selectors, Browserbase cloud runs.

Compile and optimize DSPy programs with the GEPA reflective optimizer — signatures, metrics, evolved prompts.

Instrument LLM/agent apps with Langfuse — traces, prompt versioning, datasets, LLM-as-judge evals, OTel, self-host.

Instruction-based image editing with FLUX.1 Kontext — local edits, character consistency, multi-reference prompting via BFL API.

Build sub-500ms voice AI on LiveKit Agents — STT-LLM-TTS pipelines, turn detection, telephony, MCP tools.

Drive Kiro's spec-first gate — EARS requirements, design, tasks — for traceable AWS-native migrations.

Production RAG on Postgres with pgvector — embeddings, HNSW vs IVFFlat tuning, hybrid search, scaling.

Build bi-temporal knowledge-graph agent memory with Graphiti and Zep — facts with validity windows, not flat vector recall.

Crash-proof LLM agents with Temporal — durable workflows, activities, retries, and resumable state for the OpenAI Agents SDK.

Define typed LLM functions in BAML's DSL, generate clients, and get robust structured output via Schema-Aligned Parsing.

Build production TypeScript agents with Mastra — agents, tools, workflows, RAG, and memory on Node and edge.

Evaluate RAG pipelines with Ragas — faithfulness, answer relevancy, context precision/recall, and test set generation.

Build stateful multi-step agents with LangGraph — StateGraph, checkpointing, streaming, and human-in-the-loop interrupts.