Connaught Place in Delhi has an AI Engineer opening, full time, on site, with a data science team building models that actually ship into production products rather than staying in a notebook indefinitely. Three years of hands-on experience building and deploying machine learning models is the expectation, enough to have already felt the gap between a model that performs well offline and one that holds up once real users start hitting it.
The work covers exploring and cleaning raw data, which honestly takes up more time than most people expect coming into this role, applying statistical and machine learning techniques to find patterns worth acting on, presenting findings to stakeholders in a way that doesn't require them to already understand the math, and recommending data-driven decisions that the business can actually implement. A model's accuracy score means little if nobody can act on what it found.
Required skills are Python, familiarity with common machine learning frameworks such as PyTorch or TensorFlow, deep learning fundamentals, and real experience with model deployment, not just training. A bachelor's degree in computer science, statistics, or a related quantitative field is expected, along with demonstrated hands-on experience building and deploying models using Python and common frameworks.
Connaught Place puts this team right in the center of Delhi, and the products this AI engineer's models feed into serve a genuinely large and diverse user base, which means edge cases in the data show up quickly and often in ways a smaller, more homogeneous user base wouldn't surface as fast.
Naukri Mitra has listed data science roles from this employer before, and engineers who joined in the past year describe the stakeholder communication expectation as more demanding than at previous companies they'd worked at, since this team genuinely embeds AI engineers into product discussions rather than treating the data science function as a separate, walled-off unit.
Interviews include a technical round on machine learning fundamentals and a case study working through a sample dataset, followed by a conversation about how you'd explain a model's limitations to a non-technical stakeholder. Candidates who resist the urge to oversell a model's certainty tend to interview well.
Onboarding for this role runs roughly two to three weeks, split between shadowing a current team member and working through a small, low-risk real task under review before moving to independent ownership. The team has found that a slower, more deliberate ramp-up produces fewer costly mistakes down the line than throwing someone straight into full ownership from day one.
A model that performs impressively on a held-out test set can still fail in production if that test set wasn't truly representative of real user behavior, and this team has learned, sometimes the hard way, to validate against genuinely fresh production data before fully trusting an offline evaluation number.
Model development cycles here run anywhere from a few weeks for a smaller iteration to a couple of months for a genuinely new model architecture, and engineers are expected to communicate realistic timelines to stakeholders who don't always have intuition for how unpredictable machine learning development timelines can be compared to typical software work.
The data science team holds a monthly model review session where engineers present a model's real production performance against its original offline evaluation, a practice that's surfaced more than one case where a model looked strong in testing but drifted meaningfully once real user behavior started hitting it.
This role is based at Connaught Place, Delhi, pin code 110001, five days a week on site. Pay runs up to ₹1,04,000 a month. The team hopes to fill this seat within five weeks.