Contract Period: 12 Months
Salary Range: 9000 - 12000 SGD
Location: Central
Working hours: Monday to Friday 9AM to 6.30PM
Responsibilities:
- Design multi‑agent systems for surveillance workflows using LangChain / LangGraph ,Crew AIor equivalent.
- Architect agentic RAG pipelines (hybrid retrieval, tool‑use, memory, policy control) with vector stores and knowledge graphs(neo4j or equivalent) for complex reasoning.
- Define semantic models (ontologies, entity/event schemas) and build knowledge graphs that connect trade, comms, user, and reference data.
Full‑Stack Development
- Build Python (FastAPI/Pydantic) microservices and REST/gRPC APIs that integrate LLM services, retrieval layers, and surveillance systems.
- Develop modern frontends and AI UIs using Streamlit/Gradio (and/or React if needed) for analyst workflows and explainable insights.
- Implement CI/CD pipelines, automated tests (unit/integration/load), and secure packaging/deployment across environments.
Data & RAG Engineering
- Design and operate high‑scale vector databases (Pinecone, Milvus, Weaviate) with optimal indexing strategies, chunking, embeddings, and TTL policies.
- Build data engineering pipelines for modeling, ingestion, feature extraction, and entity resolution across trades and communications.
- Construct semantic search and query planning for multi‑source reasoning.
LLMOps & Observability
- Implement LLM observability, evaluation, and monitoring (LangSmith, Arize Phoenix) including prompt/version tracking, drift detection, and error taxonomies.
- Define evaluation harnesses for faithfulness, toxicity, bias, hallucination, and task‑specific scoring (precision/recall for surveillance contexts).
- Optimize cost, latency, and reliability; benchmark OpenAI and OSS models (Llama 3, Mistral) for target tasks.
Must‑Have Qualifications
- Trade Surveillance domain experience deep familiarity with Market Abuse Regulation (MAR), FinCrime/AML, and Compliance use cases—e.g., spoofing, layering, front‑running, market manipulation, comms surveillance, alerting, and case management.
- GenAI / Agentic AI: Hands‑on with LLM integration & prompt engineering, agentic patterns, and RAG architecture end‑to‑end.
- Knowledge of vector databases and embedding strategies.
Familiarity with LangChain / LangGraph or equivalent orchestration.
- ML : LLM post-training techniques, Custom Cuda kernels for specific architectures, Distributed training infrastructure
Skills/Requirement:
- Master’s in Computer Science, Software Engineering, Data Science, or related field (or equivalent practical experience).
- 9+ years in backend/full‑stack development with Python (FastAPI, Pydantic), strong system design, testing (pytest), and API design (REST/gRPC).
- Microservices, caching, message queues, CI/CD (GitHub Actions/GitLab/Azure DevOps), and secure coding practices.
- Multi Agent System Design & Orchestration
- Full Stack AI Integration (Frontend to Backend
- Multi Cloud AI Infrastructure (AWS Bedrock, Vertex AI, Azure AI, Snowflake Cortex)
- Advanced Prompt Engineering & LLM Fine tuning, Agentic Retrieval Augmented Generation (RAG)
- Enterprise Application Integration & AI Orchestration
- Edio/Edgio & Render (Deployment & Hosting) or similar
- Cloud Platforms: AWS Bedrock, Google Vertex AI, Azure AI Studio
- Data Engineering:
- Strong SQL/NoSQL schema design, and data modeling; building scalable ingestion and transformation pipelines (batch/streaming).
- Experience with PySpark and streaming data processing (e.g., Kafka or equivalent) and feature extraction for retrieval/graph enrichment.
- APIs & Microservices: Excellent understanding of RESTful APIs and microservices patterns.
- Security & Governance: Understanding of AI safety, guardrails, PII/PHI protection, and auditability in regulated environments.
Nice‑to‑Have Qualifications
- Frameworks/Tools:LlamaIndex, semantic routers, retrieval re‑rankers; LangSmith / Arize Phoenix for LLMOps.
- Cloud: Production experience with AWS Bedrock, Vertex AI, Azure AI, Snowflake Cortex.
- Containerization/Orchestration:Docker, Kubernetes.
- Open‑Source: Contributions to AI/RAG/agentic libraries. Search & Semantics: Semantic search principles, knowledge graph tooling (Neo4j/Neptune/TigerGraph), ontology/OWL/RDF familiarity.
- Languages: Rust (for high‑performance services) alongside Python.
PERSOL Singapore Pte Ltd • RCB No. 200007268E • EA License No. 01C4394
R1551772 (Kankanala Lakshmi Prasanna)
GOREF: 15971