Sr/Lead AI/ML Engineer-Hyderabad

Job Title:                       Sr/Lead AI ML Engineer

Employment Type:        Full Time

Position Location:          Hyderabad

Qualifications:                BE/B.Tech/MCA Degree in Computer Science, Engineering, or similar relevant field

Total Experience:           8+ years

Key responsibilities

• The matcher runs in tiers. Deterministic rules go first, then string-distance and machine-learning matching, and then model-assisted classification for whatever those leave unresolved. Coverage and precision are reported tier by tier as well as in aggregate.
• The confidence score and the thresholds that decide what a person sees are part of the design. The review queue exists because a wrong match becomes a wrong price recommendation, and a client will challenge that in a meeting.
• The analyst-correction feedback loop turns overrides into signal, and it is instrumented so that quality and drift stay visible over time.
• The matching rules the analysts currently hold in their heads are written down and encoded. These include the items excluded from benchmarking, the long tail of low-volume items that is priced broadly instead of matched item by item, and the operator price-validation check.
• Point-of-sale and supplier data are queried and shaped in Compass Snowflake. Ingestion is owned by Compass and the source layer is read-only to the application, so the work sits on the read side of the pipeline.
• Mapping rules, analyst overrides and market basket versions persist in the operational store, which is proposed as Aurora Serverless v2 on PostgreSQL and awaits Compass confirmation.
• If a model is used, it is integrated through Amazon Bedrock inside the Compass account, under the AI controls Compass has set. Those controls are commitments, and they are not optional.
• The pricing-team subject-matter experts stay involved through the build. They are the arbiters of whether a match is right, and the design keeps a person in that position deliberately.

Required experience

• Advanced Python , together with the data and machine-learning libraries around it such as NumPy, Pandas and Scikit-learn.
• Production experience with text classification and entity matching is the core of this seat. That includes string-distance and similarity methods such as Levenshtein, Jaro-Winkler, TF-IDF and cosine similarity, along with vector embeddings, applied to messy real-world product or transaction descriptions.
• Complex SQL against a cloud warehouse comes up daily, and the Snowflake experience needs to cover querying schemas another team owns and controls.
• Production machine-learning practice , covering deployment, drift monitoring, closing a human feedback loop, and tracking confidence across the distribution of results.
• Experience working against a taxonomy or master catalog that changes underneath you, with versioning and a migration story, transfers directly to this seat.

Useful additions

• Experience integrating a large language model for classification or entity extraction, where the input is short, noisy commercial text and the output has to be auditable. It sits here rather than under required experience because whether the engine calls a model at all is still an open decision.
• Retail, grocery, restaurant or foodservice item data shortens the learning curve considerably. Knowing that a sixteen-ounce latte and a large latte are the same item, and that a menu description is not a product name, saves a good deal of time.
• Amazon Bedrock specifically, or any work under an AI control regime with pinned model versions and full prompt and decision logging.