AI/ML Engineer

Greetings from Javen Technologies Inc.,

Job Title: AI/ML Engineer AWS Bedrock RAG Model Location: 100% Remote Duration: 6+ Months with extensions

Required Skills (Must Have)

Technical (Core):

• AWS Bedrock handson invoking embedding & text models; experience with model customization / finetuning workflows.
• Terraform module authoring (composition, variable design, drift detection, environment promotion).
• Production RAG system experience (document processing, chunking strategies, embedding generation, retrieval optimization).
• Python engineering excellence (testable, modular code; familiarity with packaging, logging, dependency management).
• SageMaker training jobs (PyTorch estimator or equivalent, VPC config, KMSencrypted volumes & outputs).
• Vector search proficiency with OpenSearch (kNN / ANN, index design, embedding normalization).
• Step Functions (standard + distributed Map), Lambda, SQS retry patterns, SNS notifications.
• Data curation & labeling: structuring JSONL training/eval sets, metadata hygiene, dataset versioning practices.
• Retrieval & answer quality evaluation: recall@K, MRR (or similar), error categorization (hallucination vs. retrieval failure).
• Secure AWS networking: VPC subnet/AZ selection (including GPU AZ constraints), security groups, private endpoints.
• IAM & KMS usage for ML pipelines (role scoping, encryption at rest/in transit considerations).
• Observability: designing metrics/logging for pipeline latency, throughput, failure classification; CloudWatch dashboards/alarms.
• Practical prompt engineering & prompt lifecycle management (versioning, regression testing).
• Understanding of finetuning paradigms (full vs. parameterefficient (LoRA), overfitting mitigation, hyperparameter tradeoffs).

Nice to Have (Differentiators)

AWS & Platform:

• Bedrock Knowledge Bases, Agents, Guardrails early adoption / integration patterns.
• Amazon Q or similar copilots in internal tooling contexts.
• Hybrid retrieval (BM25 + vector fusion, rerankers) or experimentation with multivector approaches.

Search & ML Optimization:

• Advanced embedding strategies (domain adaptation, periodic regeneration policies).
• Index lifecycle management (ILM), hot/warm tiering, shard sizing heuristics.
• Experiment frameworks for retrieval (A/B harness, statistical significance testing).

Security & Compliance:

• Exposure to FFIEC or similar financial regulatory expectations (change control, logging, segregation of duties).
• Vault / secret management integration patterns (token renewal, secret rotation).

Data & Evaluation:

• Automated data quality pipelines (schema / semantic validation, anomaly detection).
• Prompt & answer regression harness (baseline answer store, delta classification).

Joshua Gidugu