AI Engineer

AI SSE — Agentic AI Systems
Location: Bangalore
Experience: 3–8 years
Type: Hands-on IC | Systems Builder | Production Owner

**ABOUT GOCOMET

**GoComet is a Series B logistics technology company that helps global enterprises like Unilever,
Honda, Tata, and Schneider optimize freight spend, automate procurement, and build resilient
supply chains. We're pivoting hard into named-enterprise accounts with a 4.2x growth engine.
Our product depth across procurement, planning, execution, and audit gives us a quantifiable
ROI story competitors can't match.

If you want to live inside LLMs, prompts, orchestration, and production complexity — this is your
role.

Why This Role Exists
Most teams stop at demos.
Production agentic systems fail because:
● Prompts drift
● Memory breaks
● Tools misfire
● Costs explode
● Edge cases compound
We need engineers who understand:
LLMs as distributed systems with probabilistic components.
You will make agentic AI reliable.

What You Will Own
Core Intelligence Layer
● Prompt architecture & evaluation
● Tool calling reliability
● Memory design (short-term / long-term / retrieval)
● Planning & execution loops
● Guardrails and verification
Production Systems
● Latency, cost, observability
● Failure handling & retries
● Agent debugging pipelines
● Versioning prompts & behaviors
● Continuous evaluation

Outcome Delivery

● Translate product intent into working agents
● Work with PMs to define constraints
● Ship systems that work under ambiguity
You don’t stop at: “Agent works once.”
You stop at: “Agent works reliably at scale.”

What You’ll Build
● Multi-agent architectures (planner / executor / verifier / critic)
● Tool ecosystems for agents
● Evaluation harnesses for LLM behavior
● Memory & retrieval systems
● Agent simulation environments
● Cost-aware orchestration layers
● Self-improving feedback loops

How You’ll Work
Deep LLM Engineering
● Prompt design patterns
● Structured output strategies
● Function/tool calling
● Failure mode mitigation
● Model selection & routing
● Fine-tuning / RAG / hybrid approaches
Software Engineering
● Python / TypeScript production systems
● Async orchestration
● Distributed tracing
● CI for agent behavior
● Infra for evaluation

This Is Not Just Backend Engineering
This is intelligence engineering.
You’re not wiring APIs.
You’re building systems that reason, act, recover, and improve.

Your success is measured by:
● Reliability of agent behavior
● Latency & cost efficiency
● Failure recovery
● Real outcomes delivered at scale

Debugging Intelligence
You will debug:
● Silent hallucinations
● Agent loops
● Tool misuse
● Memory corruption
● Drift across model versions
● Cost anomalies

Who You Are
You don’t ask: “What endpoint should I build?”
You ask: “Why did the agent fail on this edge case?”

Must-Have
● 4–8 years strong software engineering
● Deep hands-on LLM experience in production
● Strong understanding of:
○ Prompting strategies
○ Agent orchestration
○ Memory/RAG
○ Tool calling
○ Evaluation
● Can independently design and ship AI systems
● Strong Python or TypeScript
● Obsessed with reliability

Nice to Have
● Built multi-agent systems
● Worked on workflow automation
● Built internal copilots
● Experience with evaluation frameworks
● Experience with cost optimization
What This Role Is NOT
● Not model research
● Not API wrapper work
● Not demo building
● Not prompt hacking only

This is:
Engineering intelligence into production systems.