Principal Data Scientist – Agentic AISS&C Intralinks
About the RoleSS&C Intralinks is looking for a Principal Data Scientist – Agentic AI to provide technical leadership in designing, architecting, and delivering next-generation agentic AI systems across our enterprise product ecosystem.
This is a principal-level individual contributor role for someone who combines deep machine learning expertise, strong software engineering, and a forward-looking understanding of autonomous AI systems. You will lead the architecture and engineering of agentic platforms that can reason, plan, collaborate, use tools, remember, interact with enterprise systems, and execute complex business workflows reliably and securely.
You will work at the intersection of advanced AI research and production engineering. Beyond using existing frameworks, you will understand their internals, extend or fork them at the source-code level, and build reusable capabilities that become the foundation for enterprise-wide AI adoption. You will also influence technical direction across multiple product lines, set engineering standards, mentor senior engineers and data scientists, and drive build-versus-buy framework, model, and deployment decisions.
We operate with an agile, research-driven mindset: we continuously evaluate emerging technologies, challenge assumptions, and turn promising innovations into practical enterprise capabilities, always with a focus on reliability, scalability, security, and measurable business outcomes.
How You Will Make an Impact1. Agentic Architecture & Technical Leadership- Own the technical architecture and long-term engineering strategy for enterprise agentic AI platforms across multiple product lines.
- Establish architectural standards, reference implementations, and reusable SDKs, abstractions, and patterns for agent runtimes, orchestration, state, memory, tool integration, evaluation, and deployment.
- Select the right architecture for each use case: deterministic workflow automation, graph-based orchestration, autonomous agents, or hybrid approaches.
- Lead decisions on agent frameworks, model providers, inference infrastructure, and enterprise integration patterns, including build-versus-buy evaluations.
- Identify limitations in existing frameworks and develop custom solutions where standard implementations fall short of enterprise requirements.
- Define and communicate the long-term roadmap for agentic AI platforms, lead architecture reviews, and mentor senior and staff-level engineers and data scientists.
- Drive research initiatives, prototypes, and proofs of concept that bring emerging agent architectures, models, and protocols into production.
2. Agent Design, Planning & Multi-Agent Orchestration- Architect agents that understand objectives, decompose tasks, plan, select tools, monitor progress, and adapt based on execution results.
- Implement advanced reasoning and planning patterns: ReAct, Plan-and-Execute, hierarchical and dynamic planning, reflection, critic-verifier loops, self-correction, and consensus-based validation.
- Design supervisor-worker, hierarchical, collaborative, and decentralized multi-agent systems with dynamic delegation, sub-agent spawning, handoffs, parallel execution, and result aggregation.
- Design and operate autonomous agent swarms that self-organize around large or open-ended objectives, with coordination mechanisms for shared context, dependencies, task prioritization, resource and cost budgets, conflict resolution, failure recovery, and containment of emergent behavior.
- Build hybrid systems that combine deterministic workflows with autonomous execution so critical enterprise operations remain predictable and controllable.
- Support long-running, multi-stage workflows that maintain execution state, track intermediate results, and recover from interruptions.
- Define autonomy controls: confidence thresholds, escalation policies, human approvals, execution budgets, and bounded tool permissions.
- Make agent decisions traceable and auditable through structured plans, execution events, tool-call histories, and evidence-based decision records.
3. Agent Runtime, Harness Engineering, Skills & Autonomous Coding- Architect and extend agent runtimes and harnesses that manage the full agent lifecycle: model invocation, context assembly, tool execution, state, sessions, memory, and execution control.
- Work extensively with coding-agent environments and harnesses such as OpenCode, Claude Code, Codex-style systems, and custom environments; understand and customize the underlying execution loop, including tool discovery, context compaction, session recovery, and sub-agent coordination.
- Build custom harnesses that combine planning engines, coding agents, tool runtimes, memory systems, and workflow orchestration from different frameworks.
- Design a reusable Agent Skills ecosystem: modular skills, instructions, tool registries, and execution capabilities that are versioned, governed, and dynamically discovered and loaded based on task requirements.
- Build autonomous software-engineering workflows: repository exploration, multi-file refactoring, code generation and review, test generation and execution, debugging, vulnerability analysis, documentation, and automated remediation.
- Develop controlled execution environments (isolated file systems, shells, code sandboxes, browsers) and explore browser-use and computer-use agents for web, enterprise, and legacy interfaces.
- Optimize harness performance: context utilization, tool-selection accuracy, task completion, token consumption, and multi-step reliability.
4. MCP, Agent-to-Agent Communication & Interoperability- Lead the design of enterprise MCP infrastructure using FastMCP and compatible implementations, exposing internal applications, databases, APIs, and business capabilities as structured tools and resources.
- Define reusable tool schemas, descriptions, annotations, invocation policies, and output contracts for reliable tool discovery and execution.
- Implement remote MCP servers with OAuth-based authentication, scoped access, session management, and enterprise identity integration.
- Design agent-to-agent communication using the A2A protocol and other emerging standards (including AG-UI for streaming agent state and events), so agents built on different frameworks, models, and runtimes can discover capabilities, delegate tasks, and collaborate.
- Separate agent reasoning, orchestration, tool execution, and business logic to avoid tight coupling between agents and enterprise applications.
- Establish governance and lifecycle management for MCP tools, agent endpoints, schemas, versions, and integrations.
5. Knowledge, Retrieval, Memory & Context EngineeringAdvanced retrieval architectures
- Architect production Agentic RAG systems in which agents plan retrieval, decompose queries, choose among sources and tools, evaluate results, and iterate until answers are grounded.
- Design Cache-Augmented Generation (CAG) strategies, including KV-cache and prompt-cache preloading of stable knowledge, and define when CAG, RAG, or hybrid approaches are the right choice for latency, cost, and accuracy.
- Build GraphRAG pipelines: entity and relationship extraction, knowledge graph construction (e.g., Neo4j, memgraph), community summarization, and graph-based and multi-hop retrieval.
- Develop ontology-driven RAG that uses domain ontologies, taxonomies, and schemas to guide entity resolution, query expansion, reasoning, and answer validation for complex financial, legal, and healthcare content.
- Combine hybrid search (vector, keyword, metadata, relational) with reranking, multi-query retrieval, contextual retrieval, and query decomposition; optimize accuracy, grounding, and efficiency.
Memory & context engineering
- Lead context engineering for long-running, complex tasks: context assembly pipelines that select, prioritize, summarize, compress, and manage information within model context windows.
- Architect persistent AI memory that retains task-specific knowledge across sessions, execution boundaries, and context-window limits.
- Design multi-memory systems for agents: working, short-term conversational, long-term semantic, episodic, and procedural memory, with consolidation, summarization, temporal reasoning, entity resolution, and relevance-based recall.
- Define memory scoping across agents, users, teams, and tenants, including shared versus private memory in multi-agent and swarm settings.
- Prevent memory contamination, stale knowledge, context pollution, cross-user data leakage, and incorrect long-term retention.
6. LLM Research, Inference Optimization & Model Serving- Lead evaluation of open-source and commercial foundation models (e.g., Qwen, DeepSeek, Llama, Gemma, Muse), including reasoning, multimodal, and mixture-of-experts architectures, on reasoning quality, coding ability, instruction following, tool-use reliability, context length, latency, and cost.
- Develop model-selection and intelligent routing strategies based on task complexity, capability, latency, and cost.
- Apply model adaptation techniques: SFT, LoRA/QLoRA, distillation, preference optimization, and synthetic data generation.
- Lead inference optimization: speculative decoding, constrained decoding and structured generation, prompt and KV-cache optimization, continuous batching, quantization, and disaggregated prefill/decode.
- Design multi-node and distributed model serving: tensor, pipeline, and expert parallelism; multi-GPU and multi-node deployments; GPU scheduling and autoscaling; and cache-aware routing, using vLLM, SGLang, TensorRT-LLM, or equivalents.
- Balance quality, throughput, latency, GPU utilization, cost, and scalability across cloud, private cloud, and on-premises environments.
- Explore multimodal agent architectures spanning document understanding, vision-language models, audio, and speech.
- Establish a systematic research and experimentation process that moves promising innovations into production.
7. Agent Evaluation, Benchmarking & Observability- Architect evaluation frameworks that measure complete execution trajectories and task outcomes, not just individual LLM responses.
- Establish benchmarks for planning accuracy, tool-selection precision, invocation correctness, task completion, agent and swarm collaboration, retrieval quality (including RAG, GraphRAG, and memory recall), and execution efficiency.
- Build simulation environments and synthetic enterprise workflows for repeatable benchmarking.
- Create automated pipelines combining deterministic tests, LLM-as-a-judge, domain validation, human review, and adversarial testing, covering failure recovery, context overflow, hallucination, tool misuse, prompt injection, and inconsistent state.
- Implement end-to-end observability with Langfuse, OpenTelemetry, Arize Phoenix, W&B Weave, and custom instrumentation, tracking model calls, context usage, tool invocations, inter-agent messages, latency, tokens, cost, errors, and business outcomes.
- Establish continuous evaluation, regression testing, A/B testing, and feedback-driven improvement, with measurable quality and reliability standards for enterprise-wide deployment.
8. Production Engineering, Durable Execution & Infrastructure- Lead the engineering of highly available, scalable, fault-tolerant agent services supporting long-running tasks, asynchronous and concurrent execution, streaming, state persistence, and session recovery.
- Implement durable execution patterns: checkpointing, resumability, idempotency, retries, timeouts, circuit breakers, and compensation logic; evaluate Temporal and event-driven architectures for distributed agent workflows.
- Architect containerized services on Docker and Kubernetes, including GPU-enabled deployments and autoscaling.
- Build CI/CD pipelines, infrastructure-as-code, and reproducible deployments; manage packaging and distribution of customized frameworks, internal SDKs, MCP servers, and reusable AI components, including framework forks, dependency pinning, model versions, and environment parity.
- Drive performance optimization across orchestration, model serving, networking, databases, and distributed execution.
9. Enterprise Security, Governance & Responsible AI- Establish security-first patterns for autonomous agents operating on sensitive enterprise data: least-privilege access, agent identities, scoped IAM, OAuth, service-to-service authorization, and secure token and secrets management.
- Defend against direct and indirect prompt injection, malicious tool outputs, data exfiltration, insecure tool execution, memory poisoning, and unauthorized agent actions.
- Architect isolated execution environments with controlled access to files, shells, databases, APIs, and browsers, and enforce policy at the agent, orchestration, tool, and infrastructure layers.
- Design human-in-the-loop controls and approval workflows for high-impact, sensitive, or irreversible opera