Senior AI Engineer (Ahmedabad Candidates Only)

Senior AI Engineer

Ahmedabad

• Full-time
• On-site only

Brainy Neurals is an AI-native engineering company in Ahmedabad with nine years of shipping production AI, bootstrapped from one person to a team of fifteen-plus. We're an NVIDIA Inception member and ISO 27001 certified. Our work sits in two areas, Physical AI (camera and sensor systems that decide at the edge, like visual inspection on a production line) and Documentation AI that takes manual effort out of document-heavy business processes.

This role is for an engineer who builds those systems, not one who assembles them from whatever an AI coding tool produces. You'll train models on client data that is messy, badly lit and inconsistently labelled, then get them running inside a real workflow with accuracy numbers attached. The founder is an AI architect who started out optimizing models in C++ for edge devices, so your architecture decisions get reviewed by someone who has made them before.

What you'll do

• Build computer vision systems that go to production. Detection, segmentation, tracking and classification trained on custom client data, evaluated honestly, and deployed where someone depends on the output being right.
• Own document AI end to end. Layout analysis, printed and handwritten recognition, extraction into structured schemas, validation rules that catch bad reads, and retrieval over document sets large enough that naive chunking falls apart.
• Fine-tune and evaluate LLMs and VLMs for client-specific tasks. Build the eval harness before the model, so what you report is a measured number and not an impression from ten examples.
• Design the architecture of what you build and write it down before code starts. Data flow, model serving, API contracts, failure handling, latency and cost per inference, all of it open to review by the team.
• Read papers in your area, reproduce the parts that matter, and say plainly when a published result does not survive contact with client data. We'd rather hear that early than six weeks into a delivery.
• Ship it properly. Docker, CI/CD pipelines, cloud deployment, monitoring once it's live, and APIs that the rest of the team can integrate against without asking you how they work.

What we're looking for

• Two to three years building computer vision in production, with depth rather than breadth. You've trained on your own annotated data, debugged why a model fell apart on the client's camera angle, and taken something from a notebook to a running service.
• Real OCR knowledge. You understand why recognition degrades on skewed scans, faint handwriting, stamped and overlapping text, multi-column layouts and forms where people write outside the boxes. You know which problems belong to preprocessing, which to the model, and which to the post-processing rules.
• Theory and practice in RAG and fine-tuning, both. Chunking and embedding choices, retrieval evaluation, reranking, LoRA and QLoRA against full fine-tuning, and a clear view of when each is the wrong call. You should be able to defend these choices on a whiteboard.
• Strong Python and working C++. You can read someone else's C++ inference code and make it faster without breaking it.
• System architecture ability that goes past contributing a module. You've designed a whole system, made the tradeoffs, and lived with the consequences. Docker, cloud, DevOps practice, CI/CD and API design are part of your normal working vocabulary.
• Communication that holds up in a client room. You'll explain model limitations to a quality head who does not care about your loss curve, and you'll work alongside annotation, delivery and other engineers rather than disappearing into a branch for a month.

Preferred, not required

• Edge AI experience. Quantization, pruning, ONNX, TensorRT, Jetson or similar hardware, and the discipline of fitting a model into a power and memory budget.
• Exposure to regulated document domains such as pharma GMP documentation, insurance or banking operations, where a wrong extraction has a cost attached.

Not a fit if

• Most of your recent code came out of an AI coding tool and you cannot explain the parts you kept. We use AI tools heavily and expect you to use them too, which is exactly why we need you to know the math and the failure modes underneath what you ship.
• Your vision experience is coursework, tutorials and Kaggle on clean public datasets.
• You've only ever called a hosted API and have never trained or fine-tuned anything on your own data.
• You want remote or hybrid. This role is fully on-site at our Ahmedabad office, every working day, with no work-from-home arrangement.

What you get

• Hard problems on real client data, in vision and document AI, instead of a backlog of CRUD tickets with a model bolted on.
• Architecture reviews with a founder who has shipped edge AI in C++, and room to argue your position when you think he's wrong.
• Research time that counts as work, because our differentiation depends on us knowing the literature better than the client's other vendors.
• Direct client exposure from early on, which is how engineers here end up leading delivery.

Role details

• Location: Our Ahmedabad office, on-site, no work from home and no hybrid
• Reports to: The founder and AI architect
• Compensation: Fixed salary matched to your depth, not to your years
• Selection: A short take-home on a vision or OCR problem with your reasoning written out, then a live architecture and paper discussion where we go deep on what you submitted

How to applyEmail humans@brainyneurals.com with "AI Engineer" in the subject line, and send us these four things.

• Your resume, plus GitHub or anything else that shows code you wrote yourself.
• One vision or OCR system you built, covering the data you trained on, the accuracy you reached, what it was deployed on and where it still fails.
• One paper you implemented or reproduced, and what you learned from the gap between the paper and your results.
• Your expected CTC and your notice period.