4Bell Technology
Sr. AI Engineer Labs(R-2188)
4,000,000.00-5,500,000.00/A
Any Degree
IT (Information Technology)
Full-time
Noida
05-Dec-2026
AI Engineer Agentic AI LLM Python SQL Production AI feature AWS RAG Distributed/event-driven systems
Job Description
What You'll Do• Diagnose before building: map how the work is done today, find the constraint, and confirm the technology fits.
• Build natural-language querying over our structured and unstructured data, including structure extracted from invoices and contracts.
• Build semantic search spanning our products, delivered as a single entry point.
• Build the agentic platform: orchestration, the tool interface standard and registry, memory, budgets, retries, fallbacks, and human checkpoints.
• Close the loop on agent behavior: capture production feedback and apply the right technique — offline evaluation, preference data, bandits, or reinforcement learning — so agents improve and take on autonomous work.
• Build evaluation infrastructure: task datasets, failure taxonomies, regression suites wired into release, and LLM-as-judge calibrated against human agreement.
Design guardrails: prompt injection defense, output filtering, access control, and audit trails.
• Own model cost: routing on quality, latency, and price.
• Own initiatives end to end, through production and iteration, including a clean handoff to the owning team
What You'll Bring Required
• Carried ideas from open question to production in a small, autonomous group, often as the first or only person on the problem: framing it, building it, measuring honestly, and knowing when to stop.
• 8+ years of professional software engineering, including at least two years building AI systems that reached production and stayed there.
• At least one AI feature shipped to real users, instrumented, and measurably improved from production feedback after launch. • Distributed systems and event-driven architecture: asynchronous messaging, streams, idempotency, and eventual consistency. Our target state is event-driven.
• Retrieval: chunking, hybrid and semantic search, reranking, and measuring retrieval quality separately from answer quality.
• Prompt and context engineering, agentic patterns (tool use, planning and critic loops, orchestration), and model-choice judgment not tied to one provider's abstractions.
• Evaluating generative systems: eval sets, regression suites, and how you decided one model or prompt version was better — sample size, variance, and what the eval did not cover.
• Guardrail design: prompt injection defense, output filtering, access control, and audit trails.
• Python and SQL, and production engineering habits: tests, containers, CI, maintainable code.
• Deploying and running systems in the cloud. We are an AWS shop; equivalent depth on any cloud transfers.
• Presenting what was built, what it does not do, and what you recommend next to an executive audience.
• Bachelor's degree in a quantitative or technical field.
• Own model cost: routing on quality, latency, and price.
• Own initiatives end to end, through production and iteration, including a clean handoff to the owning team
What You'll Bring Required
• Carried ideas from open question to production in a small, autonomous group, often as the first or only person on the problem: framing it, building it, measuring honestly, and knowing when to stop.
• 8+ years of professional software engineering, including at least two years building AI systems that reached production and stayed there.
• At least one AI feature shipped to real users, instrumented, and measurably improved from production feedback after launch. • Distributed systems and event-driven architecture: asynchronous messaging, streams, idempotency, and eventual consistency. Our target state is event-driven.
• Retrieval: chunking, hybrid and semantic search, reranking, and measuring retrieval quality separately from answer quality.
• Prompt and context engineering, agentic patterns (tool use, planning and critic loops, orchestration), and model-choice judgment not tied to one provider's abstractions.
• Evaluating generative systems: eval sets, regression suites, and how you decided one model or prompt version was better — sample size, variance, and what the eval did not cover.
• Guardrail design: prompt injection defense, output filtering, access control, and audit trails.
• Python and SQL, and production engineering habits: tests, containers, CI, maintainable code.
• Deploying and running systems in the cloud. We are an AWS shop; equivalent depth on any cloud transfers.
• Presenting what was built, what it does not do, and what you recommend next to an executive audience.
• Bachelor's degree in a quantitative or technical field.