Software Development Engineer

This is a contract role, with hybrid expectations in Pleasanton, CA.

Job Summary:

Are you a seasoned engineer who thrives at the intersection of cutting-edge AI research and production-grade software engineering? This is a rare opportunity to work end-to-end on a sophisticated AI agent platform, owning everything from retrieval quality and prompt engineering to safety guardrails, structured output, and production observability. You will be a core contributor on a small, high-ownership team where your decisions directly shape the architecture, reliability, and intelligence of a customer-facing AI system. If you are energized by hard technical problems, care deeply about output quality and safety, and want to build AI infrastructure that actually works in production, this role was designed for you.

Responsibilities:

- Design and extend agent state machines, including multi-step flows with safety pre-checks, clarification loops with human-in-the-loop interrupts, tool routing, and structured output formatting
- Improve retrieval quality by tuning hybrid filtering, access-control gating, recency windows, entitlement-aware scoping, and cross-encoder reranking against a vector index
- Author and iterate on system and tool prompts using structured frameworks such as COSTAR, ReAct, and HyDE-style query expansion, with explicit hallucination guardrails and citation contracts
- Define typed schemas using Pydantic for tool inputs and agent outputs, including validators that defend against common LLM output drift such as invalid URLs, formatter leakage, and duplicate citations
- Build new agent tools that integrate with internal services including search, case creation, and recommendation systems, and route deterministically inside the agent graph
- Drive the evaluation pipeline by designing LLM-as-judge and rule-based evaluators, curating test datasets, and monitoring scheduled evaluation runs for regressions
- Wire up safety and content moderation layers for customer-facing output
- Debug production traces, instrument runs with rich metadata, and triage regressions across cascading model providers
- Partner with platform engineers on shared agent primitives and the internal model gateway

Skills:

- Fluency with modern Python 3 typing, strict static type checking with mypy, and Pydantic modeling
- Strong async Python skills including asyncio, structured concurrency, and careful handling of streaming and cancellation
- Comfort working in a large monorepo with containerized local development and modern dependency tooling
- Production experience with human-in-the-loop patterns including clarifying-question loops, agent interrupts, and resumable conversation state
- Experience designing safety and guardrail layers for customer-facing LLM products
- Familiarity with enterprise model gateways and managing provider cascades across Gemini, Claude/Bedrock, and OpenAI-compatible APIs
- Track record of iterating on prompt engineering at scale using structured frameworks rather than ad-hoc approaches
- Comfort communicating tool and response contracts to LLMs via JSON Schema and structured-output modes
- Experience sanitizing user-facing LLM output including URL validation, Markdown hygiene, and formatter-drift mitigation
- Production experience building agents with LangGraph or comparable orchestration frameworks, including state machines, conditional routing, tool nodes, human-in-the-loop interrupts, and persistent conversation state
- Strong LangChain fluency across messages, tools, structured output, runnables, retrievers, and streaming
- Experience with multi-model orchestration including provider fallbacks and cascades, and the trade-offs of streaming versus structured-output modes
- Experience streaming AI responses from LangGraph graphs to UI frameworks
- Solid grasp of vector search fundamentals including embeddings, similarity metrics, top-K selection, and similarity thresholds
- Hands-on experience with hybrid retrieval using dense and metadata filtering, and cross-encoder reranking
- Familiarity with access-control-aware retrieval that restricts results based on user identity, entitlements, or content visibility rules
- Experience with query rewriting, multi-query expansion, HyDE, and conversational context resolution
- Practical experience with LangSmith or comparable platforms including datasets, experiments, run trees, and trace debugging
- Experience building or extending LLM-as-judge evaluators and working with frameworks such as DeepEval
- Ability to design custom metrics for agent-specific behavior such as citation precision and recall or tool-selection correctness

Experience:

- 5 or more years of professional Python engineering experience in a production environment
- Demonstrated experience building and shipping LLM-powered agent systems to production
- Prior experience working on small, high-ownership engineering teams with direct responsibility for production systems
- Hands-on experience with retrieval-augmented generation systems including hybrid retrieval and reranking pipelines
- Proven experience designing and operating evaluation pipelines for LLM or agent-based systems

Qualifications:

- Bachelor's degree or higher in Computer Science, Software Engineering, or a related technical field, or equivalent practical experience
- Relevant certifications in machine learning, AI engineering, or cloud platforms are a plus

This is an exceptional opportunity to work at the frontier of applied AI engineering, where the problems are genuinely hard, the impact is immediate, and the team is small enough that your contributions are visible and meaningful. You will not be maintaining legacy pipelines or waiting on approvals to ship. Instead, you will be shaping the architecture of a production AI system from the inside, collaborating with talented engineers who care about craft, quality, and getting things right. If you want a role where deep technical work meets real-world consequence, this is it.

Tundra Technical Solutions is a global workforce and technology delivery firm, ranked by Staffing Industry Analysts as one of the largest in North America. At Tundra, we aren't just hiring top talent at the world's most recognizable brands; we are pioneers of social recruitment.

This post is for a contract assignment. Assignment duties and duration are subject to business needs and requirements. We are an equal opportunity employer, and we do not discriminate on the basis of race, religion, color, national origin, sex, sexual orientation, age, veteran status, disability, genetic information, or other legally protected characteristics.

California Notice. Qualified applicants with arrest or conviction records will be considered for employment in accordance with law, including the Los Angeles County Fair Chance Ordinance, the San Francisco Fair Chance Ordinance, and the California Fair Chance Act. Criminal history may have a direct, adverse, and negative relationship with the following job duties: protect and return property, adhere to confidentiality requirements, comply with security requirements, adhere to policies and codes of conduct.

We may use artificial intelligence tools to help our recruiters screen and assess talent. These tools do not replace human decision-making in the hiring process.

Salary : $70 - $87