# unknown

- Author: Andy
- Repository: MinhAn15/Senior-Python-Engineer
- Version: 20251213234209
- Stars: 0
- Forks: 0
- Last Updated: 2026-02-07
- Source: https://github.com/MinhAn15/Senior-Python-Engineer
- Web: https://mule.run/skillshub/@@MinhAn15/Senior-Python-Engineer~unknown:20251213234209

---

# Skill.md — Senior Python Engineer (LLM + Serverless AWS)

## 1) Role focus (from JD)
- Build production-grade LLM-powered features end-to-end: prompt design → evaluation → monitoring in production. 
- Own LLM integration layer: multi-provider routing (OpenAI/Anthropic/Cohere/Azure/Jina...), intelligent fallback, latency optimization, cost control, caching.
- Harden prompts: tool/function calling, structured JSON output for internal services (classification, generation, summarization, agents).
- Build async/event-driven services on AWS Lambda, Step Functions, SQS, EventBridge, API Gateway.
- Maintain pipelines between MongoDB, S3, vector DB (Pinecone/Qdrant/Weaviate) and internal data fabric.
- Extend Playwright-based browser automation & large-scale scraping; multimodal preprocessing (image/video → embeddings or direct model input).
- Debug LLM issues in production: hallucinations, format errors, bias, safety.
- Collaborate with product/data; contribute shared AI framework; occasionally client-facing.

## 2) Advanced Python (must-have)
### Async/await
- Patterns: concurrent fan-out (gather), rate-limit, retries, timeouts, cancellation hygiene.
- Evidence: async FastAPI/Lambda handler + integration tests.

### Typing
- Tools: TypedDict/Protocol/TypeVar+Generic to define contracts and keep clean boundaries.
- Evidence: typed provider interfaces + typed structured outputs.

### Pytest
- Patterns: fixtures, parametrization, async mocking, contract tests (schema validation).
- Evidence: tests cover parse failures + repair loop + router fallback behaviors.

## 3) LLM engineering (core)
### Structured output & prompt hardening
- Schema-first JSON via Pydantic JSON Schema.
- Repair loop for invalid JSON/validation errors (max N attempts) + logging.
- Safety: treat prompts/outputs as untrusted input; sanitize and redact.

### Multi-provider routing
- Provider abstraction + policy (latency/cost/quality).
- Circuit breaker + fallback chain + caching.
- Observability: token usage, provider latency p95/p99, fallback rate, cache hit rate.

## 4) AWS serverless / event-driven (must-have)
- Lambda: cold-start aware init, concurrency controls.
- Step Functions: Retry/Catch separation (transient vs business errors), parallel branches.
- SQS/EventBridge: buffering + routing + DLQ strategy.
- API Gateway: request validation, auth integration.
- DynamoDB Streams: event triggers (CDC style).
- CloudWatch: logs/metrics/alarms; correlation IDs.
- IAM: least-privilege roles per function/workflow.

## 5) Data pipelines (must-have)
- MongoDB as operational store; incremental extraction (CDC or periodic batches).
- S3 as staging/data lake; backfill and replay strategy.
- Vector DB upsert + metadata filters for RAG.
- Evidence: pipeline script + reindex strategy doc.

## 6) Browser automation & multimodal (must-have / nice-to-have)
- Playwright worker: queue-based execution, retries, anti-bot mitigation, DOM change detection tests.
- Multimodal: image/video preprocessing → embeddings or direct model input; batching for cost/latency.
- Nice-to-have: experience with GPT-4o/Claude 3/Gemini 1.5/Llama Vision; MCP protocol awareness.

## 7) Interview “story bullets” (talking points)
- Designed a multi-provider LLM router with fallback + circuit breaker + caching to improve reliability and control costs.
- Implemented schema-first structured outputs and a repair loop to eliminate downstream JSON parsing failures.
- Built event-driven AWS workflows (Step Functions + SQS + EventBridge) with robust Retry/Catch and production monitoring.
- Operated/observed LLM systems: traced hallucinations/format errors, added guardrails, and shipped safely.