Lead Backend AI Engineer

Overview :

TekWissen is a global workforce management provider throughout India and many other countries in the world. The job opportunity described below is for one of our clients, who has developed a core competence in creating and deploying cost-effective capabilities using an offshore-centric business model.

Position: Lead Backend AI Engineer

Location: Remote

Job Type: Contract
Duration: 6 Months

Work Type: Remote

• Seeking an elite Lead Backend AI Engineer to architect and build the core multi-agent execution runtime, domain-adapted model pipelines, and deterministic safety harnesses for our enterprise AgenticOps platform.
• In this role, you will own the engine room of autonomous IT and SecOps: translating high-level operational intent into reliable, multi-agent reasoning DAGs, grounded retrieval mechanisms, and sandboxed infrastructure actions.
• You will engineer durable agent state machines, optimize low-latency inference pipelines, train and fine-tune open-weight models for domain-specific infrastructure tasks, and construct fail-safe policy boundaries that prevent destructive executions.
• The ideal candidate pairs high-performance systems programming in Go (Golang) and Python with source-level mastery of agentic frameworks, open-source model ecosystems, and rigorous AI-driven evaluation methodologies.

Core Responsibilities:

• Autonomous Multi-Agent Runtime Engineering: Architect and implement fault-tolerant, stateful agent runtimes using Go and Python. Build advanced execution patterns including hierarchical agent-to-agent delegation, planner-executor-critic loops, multi-step reflection, and dynamic schema-constrained tool selection.
• Domain Model Training, Fine-Tuning Distillation: Design data synthesis and fine-tuning pipelines (LoRA, QLoRA, full fine-tuning, direct preference optimization) to adapt openweight models (Llama, Mistral, Qwen, DeepSeek) for specialized infrastructure tasks-such as CLI syntax parsing, telemetry anomaly correlation, and structured action generation.
• Deterministic Safety Harness Execution Guardrails: Engineer execution sandboxes, preflight dry-run validators, blast-radius limiters, and automated state-rollback mechanisms. Ensure agents cannot commit hallucinated commands or execute unverified, destructive payloads against live networks or compute clusters.
• Open-Source Agent Tooling Web-Execution Harnesses: Build extensible tool-calling layers integrating open-source agentic and computer/browser-use tooling (e.g., OpenClaw, Browser-Use, Headless CDP) alongside traditional enterprise controller SDKs, gRPC streams, and SSH/NETCONF interfaces.
• Hybrid Contextual Memory GraphRAG Substrate: Develop scalable, low-latency memory pipelines combining live topology graphs (Neo4j, AWS Neptune), vector search (pgvector, Qdrant, Milvus), and temporal session stores to provide agents with grounded infrastructure context.
• AI-Native SDLC Continuous Eval Pipelines: Implement automated, CI/CD-integrated evaluation suites (LLM-as-a-judge, synthetic scenario generation, regression benchmarks) to rigorously test agent trajectory accuracy, tool call precision, and safety compliance prior to production deployment.

Required Qualifications Technical Depth:

• Experience Track Record: 8+ years of production backend engineering in high-throughput distributed systems, with at least 3+ years specifically architecting LLM application runtimes, autonomous agent frameworks, or ML inference infrastructure.
• Dual-Language Systems Mastery (Go Python): Expert proficiency in Go (Golang) for high-concurrency, low-latency microservices, worker pools, and memory-safe network programming (goroutines, channels, context cancellation). Advanced expertise in Python (asyncio, typing, Pydantic) for AI/LLM orchestration and model training.
• Agentic Frameworks Runtime Internals: Source-code level fluency with agentic orchestration frameworks (LangChain, LangGraph, Semantic Kernel, LlamaIndex) or custom actor-model state machines. Deep understanding of how to manage durable distributed execution, state persistence across restarts, and graceful failure recovery.
• Model Training, Fine-Tuning Local Inference: Direct hands-on experience fine-tuning and distilling open-source models using Hugging Face PEFT, TRL, Axolotl, or DeepSpeed. Production expertise deploying and tuning high-throughput inference engines (vLLM, TensorRT-LLM, Ollama) and quantization formats (AWQ, GPTQ, GGUF).
• Open-Source Tooling Ecosystem: Practical experience working with open-source autonomous agent tools and execution harnesses (e.g., OpenClaw, browser automation agents, sandbox container environments like gVisor/Firecracker).
• Enterprise Infrastructure Protocol Fluency: Hands-on experience building integrations against enterprise infrastructure interfaces: REST APIs (Cisco Intersight, Catalyst, Nexus, Splunk), streaming telemetry (gNMI/OpenTelemetry), SNMP, and CLI over SSH.
• AI-Driven SDLC Evals Rigor: Proven experience establishing automated evaluation frameworks (DeepEval, Ragas, promptfoo) to continuously benchmark agent decision-making, track tool hallucination rates, and validate regression safety across model updates.
• Knowledge Retrieval Data Systems: Deep experience architecting production GraphRAG and hybrid search pipelines: dense/sparse hybrid retrieval (BM25 + embeddings), rerankers(Cross-Encoders, Cohere), and transactional data stores with graph/vector capabilities.

Mandatory:

Experience

• 8+ years of production backend engineering in high-throughput distributed systems
• 3+ years architecting LLM application runtimes, autonomous agent frameworks, or ML inference infrastructure Programming Languages
• Go (Golang): high concurrency microservices, worker pools, goroutines, channels, context cancellation
• Python: asyncio, typing, Pydantic, for AI/LLM orchestration and model training Agentic Frameworks Runtime
• Hands-on depth in at least one of LangChain, LangGraph, Semantic Kernel or LlamaIndex, or custom actor-model state machines
• Durable distributed execution, state persistence across restarts, graceful failure recovery
• Multi-agent patterns: planner-executor-critic loops, agent-to-agent delegation, schema-constrained tool selection Model Fine-Tuning Inference
• Fine-tuning open-weight models (LoRA/QLoRA, DPO) using Hugging Face PEFT, TRL, Axolotl or DeepSpeed
• Deploying and tuning inference engines: vLLM, TensorRT-LLM or Ollama
• Quantization formats: AWQ, GPTQ or GGUF Safety Execution Guardrails
• Sandboxed execution (gVisor, Firecracker or similar container isolation)
• Dry-run validation, blast-radius limiting, rollback mechanisms for destructive actions Enterprise Infrastructure Integration
• REST API integrations (Cisco Intersight, Catalyst, Nexus, Splunk or equivalent)
• Streaming telemetry (gNMI/OpenTelemetry), SNMP, CLI over SSH Evals AI-Driven SDLC
• Automated evaluation frameworks: DeepEval, Ragas or promptfoo.
• Tracking tool hallucination rates and regression safety across model updates, integrated into CI/CD

Retrieval Data Systems

• Production GraphRAG and hybrid search (BM25 + embeddings)
• Rerankers (Cross-Encoders, Cohere)
• Graph and vector stores: Neo4j or Neptune, plus pgvector, Qdrant or Milvus

Good to Have (Optional):

• OpenClaw, Browser-Use, Headless CDP
• NETCONF, gRPC streams
• Full fine-tuning and model distillation at scale
• Temporal session stores
• Experience with Llama, Mistral, Qwen or DeepSeek specifically
• Synthetic scenario generation and LLM-as-a-judge pipelines

Required Competencies

• Must possess excellent communication skills - oral and written
• Must possess knowledge of latest technology trends
• Must be a keen learner - should be able to drive Self Learning
• Must practice principle of First Time Right
• Must have an Eye for Details
• Must have high Customer Orientation
• Must be adaptable to working in multiple / matrix work environment
• Must possess good systems thinking
• Must possess good negotiation, analytical and interpersonal skills. Good leadership team player qualities.
• High on personal integrity with ability to establish relationships and work in teams and should be able to influence stakeholders. Should poses independence, robust ethics and resilience.

Educational Qualification:

• ME (IT, Computer), BE (IT, Computer), MCA, MSC-IT, BCA

TekWissen Group is an equal opportunity employer supporting workforce diversity

Disclaimer: This job posting and Location has been aggregated from external source. Role details, content, and availability are subject to change. Applicants are advised to confirm the latest information directly on the company website before applying.