AI Pipeline Engineer (Python) — Document Intelligence

Own & Extend the Production Extraction Pipeline (Python / AWS)

• Maintain and optimize our document extraction pipeline (Dockerized components + serverless architecture).

• Ensure reliable processing, correct outputs, and stable triggering of downstream jobs.

• Improve observability: logging, metrics, alerting, traceability, and cost monitoring.

Deliver New Pipelines

• Design and implement new end-to-end pipelines for new document families.

• Build evaluation datasets, quality metrics, and regression tests to prevent silent degradation.

LLM-Based Interpretation (Q&A / Structured Extraction / Risk Signals)

• Improve Q&A and interpretation quality using LLMs.

• Implement structured outputs, retrieval (RAG), deterministic validation, and post-processing.

• Build robust failure handling (timeouts, retries, fallbacks, safe defaults).

GenAI Foundations

• Implement “agentic” building blocks in Python behind stable APIs.

• Collaborate with product engineers to integrate AI workflows into the platform.

Production Readiness

• Ensure pipelines are secure, scalable, and cost-efficient.

• Contribute to deployment practices, versioning, rollbacks, and environment management.

What We’re Looking For

Must-Have

• Strong production-grade Python (APIs/services, testing, packaging, clean architecture).

• Experience owning and maintaining production systems.

• AWS serverless experience, especially Lambda + S3 (bonus: Step Functions, SQS, CloudWatch).

• Docker and experience operating containerized components.

• Proven experience debugging and maintaining automated pipelines in production.

• Practical experience with LLMs (OpenAI/Azure OpenAI/Anthropic):

- Structured outputs

- Prompt iteration

- Retrieval (RAG)

- Evaluation approaches

Nice-to-Have

• Experience with document AI (OCR, layout/text extraction, noisy PDFs).

• Evaluation-driven development (test sets, regression checks, quality metrics, cost/latency budgets).

• Familiarity with TypeScript/Node and/or agent orchestration frameworks.

• Experience integrating with RESTful services.

First 90 Days – What Success Looks Like

Weeks 1–2: Takeover & Stability

• Set up local + staging environments for the existing pipeline.

• Document architecture, failure modes, and operational playbooks.

• Establish baseline observability (logs/metrics).

• Define quality and cost targets (latency, cost per document, success rate).

Weeks 3–6: New Pipeline Delivery

• Deliver a new end-to-end pipeline for a new document family.

• Create evaluation datasets and automated regression checks.

• Ship a production-ready version with monitoring and rollback strategies.


Weeks 7–12: Quality, Scale & GenAI Foundations

• Improve extraction/Q&A accuracy through structured outputs, retrieval, and validation.

• Deliver an initial GenAI “kernel” API (planner/drafter/risk checks) ready for orchestration.

• Harden operations: cost controls, retries/fallbacks, production runbooks.

Tech Stack

• Frontend: React (TypeScript)

• Backend: Java (JEE)

• AI / Document Processing: Dockerized OCR/NLP components, AWS Lambda-based Python pipelines, AWS S3

• LLMs: Azure-hosted ChatGPT / LLM APIs

• Infrastructure: AWS + Azure (hybrid)