Senior AI & Document Intelligence Lead - Responsible AI (PL5550)
Paralucent
- Toronto, Canada
- Hybrid
- Lead/Staff
- 8+ years
- Posted
Skills
Required
- OCR / document intelligence / NLP / LLMs / retrieval-grounded enterprise applications
- error analysis
- machine learning
- applied AI
- data science
- document processing
- evaluation design
- human-in-the-loop controls
- AI observability
Preferred
- AI
- Responsible AI / cloud AI / machine-learning certification
- workflows
- Azure
- OpenAI
- AWS AI services
Paralucent is seeking a Senior AI & Document Intelligence Lead to design and validate AI-assisted capabilities for high-stakes, regulated workflows. You will lead OCR, classification, metadata extraction, summarization and recommendation use cases while ensuring source grounding, confidence controls, human validation, privacy, security, traceability and reliable fallback.
Responsibilities
- Define the AI/document-intelligence architecture, providers, models, orchestration, data boundaries and deployment options.
- Design OCR, document classification, metadata extraction, language detection, completeness checks, summarization and recommendation capabilities.
- Create representative evaluation datasets covering languages, document types, scan quality, ambiguity and edge cases.
- Define measures such as precision, recall, classification accuracy, confidence calibration, false negatives, unsupported outputs and human acceptance.
- Implement source grounding, model/version logging, prompt/template governance, review/override flows and fallback to human processing.
- Partner with business and legal/control stakeholders to define permitted and prohibited AI actions.
- Ensure sensitive information, prompts, embeddings, retrieval content and logs are handled according to approved privacy/security requirements.
- Support Responsible AI reviews, model changes, monitoring, drift/error analysis and incident response.
- Clearly distinguish production-proven, configurable, developmental and experimental capability.
- Coach engineering and QA teams on safe implementation and AI-specific testing.
- Success Looks Like AI use cases are selected based on measurable operational value and risk, not novelty.
- Evaluation evidence supports every production recommendation and known limitations are visible.
- Low-confidence, ambiguous and sensitive outcomes reliably route to qualified human validation.
- AI outputs remain traceable to source content and human decisions.
- Models can be changed, monitored or disabled without disrupting the controlled workflow.
- Understanding of privacy, security, model risk, data residency, retention and audit requirements.
- Illustrative Measures Final targets and cadence will be confirmed for the specific assignment.
Evaluation coverage and quality
- Closes loops, communicates risks early and remains accountable across handoffs.
- Clarifies the real problem and proposes practical options with trade-offs.
- Understands operational reality, success measures and the impact of decisions on users and the institution.
Requirements
- This is not an experimentation-only role; it requires disciplined movement from evaluation to controlled production use.
- 8+ years across machine learning, applied AI, data science or intelligent document processing, with senior technical ownership.
- Demonstrated production experience with OCR, document intelligence, NLP/LLMs or retrieval-grounded enterprise applications.
- Strong knowledge of evaluation design, confidence calibration, error analysis, human-in-the-loop controls and AI observability.
- Ability to explain AI trade-offs and limitations clearly to executives, control functions and engineering teams.
- Brings sound judgment, explains reasoning and actively invites challenge and feedback.
Preferred
- Experience in banking/financial services preferred; regulated-industry AI experience accepted.
- Experience Experience with Azure AI/OpenAI, AWS AI services or comparable enterprise platforms.
- Experience with English/French document processing.
- Experience with legal, compliance, investigations, disclosure or document-heavy operational workflows.
- Responsible AI, cloud AI or machine-learning certification.
Production acceptance of AI use cases False-negative/unsupported-output trends Human acceptance and override patterns Confidence calibration and fallback performance Model monitoring and issue closure Responsible AI evidence readiness How You Demonstrate Our Core Values Driven: Moves important work forward, recovers quickly from blockers and focuses on meaningful outcomes.
