Agentic Forward Deployed Engineer (Mid Level)

Mechanized AI Consulting Group; India; Remote; INR 8,571,000 - 12,381,000; Full time

Job summary

Mechanized AI Consulting is seeking an experienced software engineer to own client deployments of agentic systems. You will design, build, and deploy production agents inside the client environment, coordinating with stakeholders and ensuring measurable results.

You will work in a forward-deployed role across US, India, or LATAM time zones, and you will speak conversational English in client meetings. You will ship safe, scalable solutions with a focus on reliability.

Qualifications

• 3–5 years of professional software engineering (AI preferred).
• Shipped at least one hard production agent or agentic system.
• Forward-deployed, customer-embedded technical delivery.
• Conversational English; you run the client meeting and write design notes.
• Strong Python skills; experience with LangGraph, AutoGen, AG2, LlamaIndex, CrewAI, or equivalent.
• Comfortable in a client VPC or private landing zone.
• Ability to overlap US working hours from US, India, or LATAM.

Responsibilities

• Sit with critical client stakeholders and name the real problem, constraints, and the smallest proof that would change their mind.
• Do forward-deployed, customer-embedded technical delivery: you are in the client’s work, not adjacent to it.
• You run the client meeting yourself, in conversational English.
• Leave a handoff the client can run without you: runbooks, evals, traces, and the next change the platform must make.
• After the first thing is used, pull the next problems out of the client’s actual work. Scope the smallest next ship.

Skills

Python

LangGraph

AutoGen

LlamaIndex

CrewAI

AG2

Conversational English

Tools

APIs

Job description

Contract · 6–12 months, extension possible · Remote (US, India, or LATAM)

About Mechanized AI Consulting

Mechanized AI Consulting builds agentic workflows that drive real business value for enterprise clients. We sit with the client, name the hard problem, and leave a production path — not a demo.

This is the front of applied agentic systems. The work is client-facing. The output is a deployment the client can run.

The Role

You own a client deployment from the real problem to a system they can run without you. Production agents that take actions in the client’s systems (tools, data, identity) are the usual product. You own the skills, the MCPs, the RAG, the data loading, and the agent itself.

Once the first deployment is in use, you find the next real problem in the account, name it, and ship that too. You work in the client’s systems and meetings, not their office. You think in systems. You hunt failure modes before they show up in a trace. You report to the Chief AI Officer. You own the build. The Chief AI Officer owns the account and the hard no.

Overlap with US working hours is required.

What You Will Do

Client delivery

• Sit with critical client stakeholders and name the real problem, the constraints, and the smallest proof that would change their mind.
• Do forward-deployed, customer-embedded technical delivery: you are in the client’s work, not adjacent to it.
• You run the client meeting yourself, in conversational English.
• Leave a handoff the client can run without you: runbooks, evals, traces, and the next change the platform must make.
• After the first thing is used, pull the next problems out of the client’s actual work. Scope the smallest next ship.

Agent design and build

• Design the agent: skills, tools, memory, state, and the control loop.
• Stand up MCP servers and tool scopes with least privilege.
• Build retrieval the client can trust — RAG, graph retrieval where it fits, and the data-loading path into both.
• Load, clean, and partition messy enterprise data so the agent can actually use it.
• Wire the agent to the client’s APIs, files, identity, and event streams.

Systems thinking

• See the whole path: data → loading → retrieval → tools / MCP → skills → agent → eval → the client’s outcome.
• Name the contract between each part. What must be true. What is allowed to change.
• Ask what happens when a part lies, is slow, is empty, or drifts — before it happens.
• Design the fail-closed path in advance. Know what can change independently without breaking the rest.

Scientific practice

• Hypothesize failure modes up front: wrong retrieve, tool over-scope, schema reject, silent miss, cost blow-up.
• Put a check or a bound in before production. Do not wait for a trace to tell you the system is broken.
• Traces confirm a design. They do not replace one.
• Measure groundedness, latency, cost, and success rate in a way the client can repeat.

Production

• Ship the agent into an environment other people run — the client’s cloud path (AWS, Azure, or GCP).
• Instrument quality, cost, and latency.
• Build safety in from the first tool call: OWASP LLM Top 10, scoped tools, output validation, fail-closed defaults.

What We Are Looking For

• 3–5 years of professional software engineering (AI preferred).
• You have shipped at least one hard production agent or agentic system that other people actually run.
• Forward-deployed, customer-embedded technical delivery. This is required. You have sat inside a customer’s work and shipped in their environment.
• Conversational English. You run the client meeting and write a clear design note in the same week.
• Strong Python.
• Hands-on with LangGraph, AutoGen, AG2, LlamaIndex, CrewAI, or equivalent.
• Comfortable in a client VPC or private landing zone.
• Able to overlap US working hours from the US, India, or LATAM.

Nice to Have

• You have turned a first production win into a second use case in the same account.
• Graph retrieval / GraphRAG.
• Observability: Langfuse, Phoenix, or OpenTelemetry.
• IaC: Pulumi, Terraform, or Cloud SDKs.
• SSO / IdP, allow-lists, and data residency.