Kuku Technologies logo

AI Engineer

Kuku Technologies

Mumbai
0-1 Years
Full-time
₹15 – ₹28 LPA
Posted 50 mins ago
Kuku Technologies Banner

Kuku Technologies, the team behind Kuku FM and Kuku TV, is hiring an AI Engineer in Mumbai to build systems that turn raw source material into finished audio and microdrama content at scale. The role sits between engineering and storytelling: you will design prompt pipelines, orchestrate multi-step agentic workflows, deploy open-source generative models, and write evals that measure whether output is genuinely good, not just technically valid. Your work feeds directly into content that reaches a large multilingual audience, so judgment and craft matter as much as code.


Most AI roles ask you to make a model work. This one asks you to make a model’s output worth listening to, which is a harder and more interesting problem. Kuku Technologies is looking for an engineer who can build production pipelines and also tell when a generated story beat falls flat.


Why This Role Matters

Kuku operates one of India’s largest storytelling platforms, with audio stories, short courses and microdramas across several languages. Producing that volume by hand does not scale, so the company is building AI-assisted production pipelines. As an AI Engineer, you become part of the machinery that converts source text into finished creative output. A pipeline that is slightly off does not just fail a test, it produces content that feels hollow to listeners. That is why the job description stresses taste alongside technical strength.

On-site Mumbai Full-Time


What You Will Actually Build

The work spans the full lifecycle of a generative content pipeline.

Area

What it involves

Prompt pipelines

Designing, testing and iterating multi-stage prompts and agent components

Agentic workflows

Orchestrating tool-using agents that plan, call models and hand off between stages in production

Model serving

Setting up inference endpoints, managing GPU compute, tuning for latency and cost

Evals

Comparing outputs against references, quantifying quality, feeding results back into the pipeline

Data pipelines

Structured extraction from messy text, localization, similarity checks and deduplication

Production support

Debugging and maintaining live pipeline stages


A Typical Day In This Role

Expect to move between building and judging. A morning might go to debugging a pipeline stage that stopped handing off cleanly to the next agent. Later you could review a batch of generated scenes against reference material, adjust a prompt, and rerun your eval suite to confirm the change actually improved quality instead of just changing it.

Because the output is creative, you will often review results by reading or listening, not only by checking metrics.

Other days lean toward infrastructure: spinning up an inference endpoint for an open-source model, checking GPU utilization, or reducing cost per generation without hurting quality.


Technical Skills Worth Having

Here is how the requirements break down between what is expected on day one and what will set you apart.

Must have

Strong advantage

Solid Python with clean, readable code

Agent frameworks such as LangGraph LangChain CrewAI

Hands-on use of LLM APIs and prompt engineering

Vector databases and RAG, for example Qdrant pgvector

Comfort with Git , REST APIs and Linux

Open-source media models like Flux, LTX or Wan

A feel for content and narrative

Serving models with vLLM , ComfyUI, Cog or Docker with GPU

Clear communication

Evals and LLM-as-judge scoring, plus FastAPI, Node.js or AWS

Personal projects count as LLM experience


Who Can Apply

The posting asks for roughly 1 to 3 years of engineering experience, or a strong portfolio of shipped projects. That second option is significant for freshers. A recent graduate with a working multi-step LLM pipeline, a deployed model endpoint or a documented eval harness can compete with candidates who have a few years of experience. Visa sponsorship is not available, and relocation is allowed for candidates moving to Mumbai.

A shipped project that shows judgment about output quality will say more than a long list of libraries.

Things Worth Learning Before Applying

If you want to improve your odds, spend time on these areas first.

Start with agent orchestration. Build a small workflow where one agent plans, another calls tools, and a third reviews the result. Notice where handoffs break, because that is where production problems live.

Next, learn evaluation. Write a simple harness that scores generated text against references and try an LLM-as-judge setup. Be ready to explain where automated scoring misleads you and why a human read is still needed.

Finally, get one open-source model running behind an endpoint, even a small one. Knowing how GPU memory, batching and latency interact is useful in an interview and on the job.


Content Sense and Craft

This is the unusual part of the role. The team wants someone who can distinguish output that is merely correct from output that is actually good. In practice that means noticing flat dialogue, weak pacing, inconsistent tone, or a translation that is accurate but loses the feel of the original. Reading widely, listening to audio fiction, and watching short-form drama will sharpen that instinct, and you should be prepared to discuss examples during interviews.

Taste


What Recruiters May Evaluate

Expect interviewers to look at your Python code quality, how you structure a prompt pipeline, and how you would measure improvement. You may be asked how you would reduce inference cost, handle a failing pipeline stage, or compare two prompt versions fairly. Communication also counts: the posting values candidates who ask good questions instead of staying stuck silently.

flowchart LR
A[Portfolio Review] --> B[Technical Round]
B --> C[Pipeline Design Discussion]
C --> D[Final Round]

Career Direction From Here

This role builds a rare mix of skills: production LLM engineering, model deployment, evaluation, and creative judgment. Over time it can lead toward senior AI engineering, applied research in generative media, or ownership of content-automation platforms. Experience running agentic systems in production for a consumer entertainment product is also valuable in the wider generative AI market.


Keywords for Resume

Python • LLM APIs • Prompt Engineering • Agentic Workflows • LangGraph • LangChain • CrewAI • RAG • Vector Databases • Qdrant • pgvector • Embeddings • vLLM • Docker • GPU Inference • FastAPI • REST APIs • Git • Linux • AWS • LLM-as-Judge • Evals • Generative AI


Final Thoughts

This is a strong fit for engineers who want their work to be heard and watched by real audiences. If you can build reliable pipelines and also care whether the result is good, the role offers hands-on exposure to generative media at production scale.


The above article is written by me, a person interested in technology, automobiles, modern gadgets, movies, music, and clean aesthetics.

Disclaimer

This job listing is shared for informational purposes only. We are not affiliated with the hiring company. All applications must be submitted through the official company website.

Recent Postings

JJOBS