PhonePe AI Fresher Track: Fraud, Ranking and Recommendations

PhonePe posted an entry-level Data Scientist role explicitly targeting freshers in April 2026, and its ML systems run at some of the highest UPI transaction volumes in India.

That combination of fresher eligibility and hyperscale production is rarer than it sounds. Most companies operating at that transaction scale expect two to three years of industry ML experience before they put a new hire anywhere near a fraud model. PhonePe’s April 2026 posting confirms this is not rare by accident. The company is expanding its AI surface area fast enough to onboard new graduates directly into production ML teams, as confirmed via its Greenhouse careers system.

This article covers each ML domain, the selection process, role tracks, and what to have ready before applying.

Why PhonePe’s AI Scale Is Different

PhonePe is Walmart-backed (following the Flipkart spinoff in 2023) and handles a substantial share of India’s UPI transaction volume. Running ML on that infrastructure is not a prototype environment. A fraud model in production at PhonePe makes real-time decisions on financial transactions at volumes that most enterprise ML teams in India never encounter in their careers.

That scale changes what the fresher role actually demands. The team already knows ML works on payments. The open question when they interview a fresher is whether the candidate can write production-quality feature pipelines, reason about false positive rates at scale, and understand the cost difference between a missed fraud flag and a blocked legitimate transaction. Theoretical knowledge of classification does not answer those questions. A project that shows you have grappled with them does.

PhonePe’s engineering campus is in Bangalore, and all three fresher-eligible tracks (SWE, Data Scientist, and Android SWE for Indus AppStore) are Bangalore-based. Remote options are not listed on current postings.

The Four ML Surfaces You Will Actually Work On

PhonePe’s official careers page lists six engineering verticals: Core UPI Payments, Data Platform, Pincode, Indus AppStore, Financial Services, and Security Engineering. The ML and Data Science work sits across four of these, and each has a distinct problem shape.

UPI Fraud Detection

Real-time transaction risk scoring is the highest-stakes ML surface at PhonePe. A fraud model here makes a classification call in milliseconds. The engineering constraints are tight: low latency, high throughput, and a severe class imbalance problem because fraud events are rare relative to total transaction volume. Standard accuracy metrics are close to meaningless in this setting. Precision at a given recall threshold, false positive rate, and the operational cost of each error type are what the team actually tracks. Classification, anomaly detection, and threshold tuning are the core ML skills here, not deep learning for its own sake.

Financial Product Recommendations

PhonePe’s financial services vertical covers insurance and lending products embedded in the app. The recommendation engine is a collaborative filtering or contextual bandits problem, but the dataset is large and financial regulation around what can be recommended to whom adds a compliance layer that most recommendation tutorials skip entirely. Python, SQL, and familiarity with A/B testing for recommendation quality are the relevant skills.

Indus AppStore App Ranking and Discovery

The Indus AppStore supports 12 regional Indian languages. PhonePe launched the Indus AppStore Developer Platform with zero listing fees in year one and no commission on third-party payment gateway purchases. The ML problem in discovery and ranking involves multilingual content understanding, install and engagement signals across a non-English-dominant user base, and cold-start handling for newly listed apps. The developer base is growing, which means fresh data and more edge cases than an established store.

Pincode Hyperlocal Demand Forecasting

Pincode is PhonePe’s hyperlocal commerce product. Demand forecasting at pin-code granularity (rather than national aggregate) is closer to operations research than to NLP or vision: time series modeling, location data, and inventory signals combined into a short-horizon forecast that a local merchant can act on. This surface is less prominent in fresher postings but is worth knowing if you have done time series work.

The Selection Process, Step by Step

Per PhonePe’s documented recruitment process, the hiring sequence for engineering roles runs across five stages:

• Stage 1 — Coding assessment: Aptitude and coding problems. Data Scientist candidates can expect Python and SQL focus. Indus AppStore SWE candidates can expect Android and Java/Kotlin focus. DSA fundamentals covering arrays, graphs, and dynamic programming are the core test material across all tracks.
• Stage 2 — Technical interview, Round 1: DSA problem-solving under live observation. Expect coding in your language of choice. The interviewer tests both the correctness of your solution and the way you break down the problem before you start typing.
• Stage 3 — Technical interview, Round 2: System design. For Data Scientist and ML roles, expect questions about ML pipeline architecture, feature stores, model serving, and monitoring in a streaming environment. For SWE roles, expect standard distributed systems design questions.
• Stage 4 — Hiring manager round: Discussion of the specific team’s work, your project experience, and how you approach open-ended technical problems. Having a well-articulated view of one of PhonePe’s four ML domains strengthens this round considerably.
• Stage 5 — HR round: Compensation discussion, joining timeline, and relocation logistics for candidates outside Bangalore.

The stated minimum aggregate per PhonePe’s recruitment guide is 60%, applying across all three fresher-eligible tracks. For Stage 1 and Stage 2 preparation, the most-asked data structures interview questions covering arrays, graphs, and dynamic programming form the practical drill set that maps directly to PhonePe’s stated test material.

One distinctive aspect of the April 2026 Data Scientist posting: ML project portfolio weighted heavily in the resume screen. That is unusual for a fresher role, where competitive GPA and coding assessment score typically dominate. Portfolio weighting means a deployed ML project with documented results carries real weight against candidates who have only coursework to show.

Building the Portfolio That Gets You Through the Screen

The 2026 AI Roadmap for Indian Engineering Students covers the curriculum path in detail, including free resources for building ML fundamentals from scratch. What it also makes clear: two deployed projects on a public GitHub beat any stack of certificates when an interviewer wants to know what you have actually shipped.

For a PhonePe Data Scientist application, three project types map directly to the four ML surfaces above:

• A binary fraud classifier trained on a public dataset, such as the IEEE-CIS Fraud Detection dataset. The dataset is widely used, but it works if you can explain every pipeline decision, discuss your precision-recall trade-off, and describe what you would change in a production setting with live data arriving as a stream.
• A collaborative filtering or content-based recommendation engine with documented precision and recall at K values. The insurance recommendation context at PhonePe means interpretability matters alongside raw metric performance.
• A time series demand forecast for a small real dataset. Public retail sales data or a self-collected dataset from a local market both work. The goal is to show you can frame a forecasting problem correctly, not to demonstrate access to proprietary data.

None of these requires a GPU cluster. They require Python, scikit-learn or PyTorch, SQL for feature engineering, and a public GitHub repository with a clear README covering the problem, the approach, and the evaluation results.

The quality bar in a PhonePe interview is not “did you build this” but “can you explain every choice you made and what you would change given more time, more data, or a tighter latency constraint.” That is the conversation the hiring panel wants to have.

Building these projects means making real API calls, handling messy data, and running into the errors that tutorials skip. TinkerLLM is where you move from reading about LLM-backed feature engineering to doing it. For ₹299 you get live LLM API access to build a data-enrichment layer that mirrors what a PhonePe fraud team would actually use, and that project is what you bring to the interview when the panel asks what you have shipped.

Role Tracks and Salary Bands

Three fresher-eligible tracks are active at PhonePe as of 2026. PhonePe does not publicly disclose CTC figures. The ranges below are aggregator estimates from job posting data and are not official PhonePe disclosures.

Role	Experience	Est. CTC (aggregator estimate)	Core Stack
Software Engineer (SDE-1)	0 to 3 years	~28 LPA	Java or Kotlin, Python, distributed systems
Data Scientist	0 to 2 years, fresher eligible	26 to 37 LPA	Python, SQL, ML fundamentals, A/B testing
Android SWE, Indus AppStore	0 to 3 years	22 to 35 LPA	Android, Java or Kotlin, app store ML

All three roles are based in Bangalore. The Data Scientist band is wider than the SWE band because the upper end reflects ML specialisations in fraud or financial services that command a premium even at entry level.

For the SWE track, Java or Kotlin is the preferred language for the payments stack, per PhonePe’s published role preferences. Python is the preferred language for the Data Platform and Data Scientist tracks. CSE or IT freshers who have done most of their coursework in C++ or Java should expect to spend time before the application window working through Python, pandas, and SQL at a working-professional standard.

ECE freshers are not excluded. The Indus AppStore Android SWE track and the Pincode data roles have posted requirements that ECE candidates with strong programming foundations can meet. The same 60% aggregate applies uniformly across all tracks and branches, per PhonePe’s published recruitment guide.

The SWE role on PhonePe’s Greenhouse job board lists the 0-to-3-years bracket explicitly, confirming that fresher and early-career engineers are in scope for the core payments engineering track, not only the dedicated Data Scientist opening.