AI Architect
No salary listed
Expert
Chennai, Tamil Nadu, India
HybridHybrid work in Chennai.
AI & Machine LearningNatural Language Processing & LLMs
About the job
Requirements
- Architected and operated at least one LLM agent system with real users and production traffic, including tool calling, multi-step workflows, failure handling, and human escalation, and owned it after launch through incidents, cost, and latency.
- Deep knowledge of encoders and transformers, including attention, positional encodings, encoder-only, decoder-only, and encoder-decoder architectures, tokenization, and the KV cache.
- Trained or fine-tuned encoder models such as BERT-family models, sentence embeddings, bi-encoders, and cross-encoder re-rankers for classification, extraction, or retrieval; understand how contrastive training, distillation, and quantization affect cost and quality.
- Experience managing token budgets and building retrieval and memory for agents; used compaction and sub-agents to maintain coherence on long tasks and measured quality changes as context windows fill.
- Built model execution layers including agent loops, tool dispatch, state and checkpoints, interruption and recovery, sandboxing and permissions, guardrails, and model routing; fluent with MCP, structured outputs, and streaming.
- Built offline evaluation suites including datasets, graders, LLM-as-judge calibration, and CI regression gates, as well as online production scoring, trace sampling, drift monitoring, shadow deployments, and feedback loops; used evaluations to block releases or justify model changes.
- 10+ years building software or ML systems, including 3+ years building on LLMs.
- Strong hands-on Python skills and ability to read and review Java or TypeScript services.
- Production experience with AWS or GCP, containers, Kubernetes, and event-driven systems.
- Ability to explain technical trade-offs to engineers and their business impact to executives.
Responsibilities
- Define the reference architecture for FourKites agents, including the agent loop, tool execution, state and checkpointing, retries and recovery, guardrails, permissions, and human-in-the-loop handoffs.
- Design agent systems for long-running, multi-step work spanning multiple context windows and recoveries from tool, carrier API, or model-call failures.
- Choose whether problems are best handled by frontier LLMs, open-weight models, fine-tuned encoders, or code; own model routing, cost-per-task, and latency budgets across the fleet.
- Integrate agents with FourKites' event-driven platform, including Kafka, microservices, and Postgres, and with customer systems through APIs and MCP servers.
- Set standards for agent context windows, including system prompts, tool definitions, retrieved documents, memory, and prior turns, within explicit token budgets.
- Build shared components for retrieval, compaction and summarization, sub-agent isolation, prompt caching, and persistent memory.
- Diagnose context failures such as lost instructions, tool confusion, and quality degradation as context windows fill, and turn fixes into reusable harness patterns.
- Build the offline evaluation framework, including golden datasets from production traces, code-based checks, LLM-as-judge graders calibrated against human labels, and CI regression suites that gate prompt, model, or tool changes.
- Build online evaluation through scoring sampled production traffic, trace- and task-level quality metrics, drift detection, shadow and A/B rollouts, user feedback signals, and quality alerts.
- Connect evaluation metrics to customer business outcomes such as tracking accuracy, exceptions resolved without a human, and appointment fill rate.
- Review agent-team designs and pull requests, write architecture documents and playbooks, and mentor staff and senior engineers.
- Track model and tooling developments, run relevant experiments, and give leadership clear build, buy, or wait recommendations.
- Partner with Product, Solutions, and Customer Success to turn field failures into platform fixes.
Desired Qualifications
- Experience in supply chain, logistics, or transportation.
- Document AI experience, including OCR, layout models, and extraction from bills of lading, invoices, and PODs.
- Experience building voice or email agents that act on behalf of a business.
- Agent security experience, including prompt-injection defense, tool permissioning, and tenant data isolation.
- Hands-on experience with multiple model families such as Claude, GPT, Gemini, and open-weight models, and with serving stacks.
- Published work, talks, or open-source contributions on agents, retrieval, or evaluation.
About the company
FourKites provides a data-driven platform for real-time visibility and predictive analytics across large, end-to-end supply chains. It ingests data from millions of global shipments daily and uses patented AI to turn that data into actionable insights, with Fin AI offering a natural language interface to spot optimization opportunities and automate tasks. Its platform links every point of the supply chain so teams share a single source of truth and can respond faster. Its goal is to help brands reduce risk, improve responsiveness, and optimize performance across their networks.
Company Size
501-1,000
Company Stage
Late Stage VC
Total Funding
$211.6M
Headquarters
Chicago, Illinois
Founded
2014
Get referred to FourKites
See people who can refer or advise you
Simplify's Take
What believers are saying
- On September 10, 2026, Nucleus named FourKites a Leader in control towers.
- FourKites says 500 workflows and 13.8 million customer actions run in production.
- Loft launched February 9, 2026, expanding sellable AI across ERP, WMS, and CRM.
What critics are saying
- Project44 litigation still shadows FourKites after Illinois Supreme Court proceedings in 2024.
- Descartes MacroPoint, project44, and Shippeo compress pricing as visibility tools commoditize.
- A failed autonomous booking or inventory transfer would kill trust and freeze enterprise rollout.
What makes FourKites unique
- FourKites pairs real-time network data with digital twins across shipments, inventory, and orders.
- Loft turns SOPs into governed automations, giving FourKites sticky workflow depth beyond tracking.
- Booking Connect for Ocean closes the loop from visibility to autonomous freight execution.
Help us improve and share your feedback! Did you find this helpful?
Benefits
- Medical, dental, & vision
- PTO
- 401k with company match
- Annual wellness stipend
- Ongoing learning & development opportunities