profile-pic

Subhankar Halder

I work as an AI/ML Developer at HSBC India, where I architect and build ML platforms that power financial trading strategies. I design scalable inference systems, orchestrate GPU compute infrastructure, implement model monitoring solutions, and evaluate cloud ML offerings. My work spans the entire ML lifecycle—from infrastructure design to production deployment and monitoring.

  • Role

    Software Development & Deep learning Engineer

  • Years of Experience

    6.67 years

Skillsets

  • Python
  • machine_learning
  • NLP
  • PyTorch
  • FastAPI
  • Kubeflow
  • Django
  • DVC
  • HuggingFace
  • MLFlow
  • OpenCV
  • pandas
  • Scikit-learn
  • Streamlit

Professional Summary

6.67Years
  • May, 2024 - Present1 yr 11 months

    AVP AI/ML Developer

    HSBC
  • Nov, 2022 - May, 20241 yr 6 months

    Software Development Engineer

    Fynd
  • Apr, 2022 - Nov, 2022 7 months

    Deep Learning Engineer

    Paralaxiom Technologies Private Limited
  • Jul, 2015 - Oct, 2015 3 months

    Analyst

    Arcesium
  • Dec, 2019 - Mar, 2020 3 months

    Intern

    Paralaxiom Technologies Private Limited
  • Mar, 2020 - Mar, 20222 yr

    Associate Software Engineer

    Paralaxiom Technologies Private Limited
  • Apr, 2015 - Jun, 2015 2 months

    Analyst

    D. E. Shaw India Private Limited

Work History

6.67Years

AVP AI/ML Developer

HSBC
May, 2024 - Present1 yr 11 months

Software Development Engineer

Fynd
Nov, 2022 - May, 20241 yr 6 months

    - Finetuned Llamav2 LLMs for Jio Copilots Showcase at Meta Connect 2023 Conference

    - Finetuned Llamav2 LLMs on A100 GPUs to enhance Jio Copilots chat & intent capabilities

    - Employed PEFT, LORA and Supervised Fine Tuning (SFT) techniques to finetune LLMs

    - Maintained BLEU & ROUGE scores while qualitatively evaluating the trained LLMs efficacy

    - Converted LLMs into quantized CPU models using Llama.cpp while storing models in HuggingFace

    - Deployed LLM models on GPUs using HuggingFace Endpoints, FastAPI and Streamlit

    - Engineered Image Based Semantic Search leveraging CLIP model, PyTorch and Milvus DB

    - Utilized OpenAI Clip, PyTorch, and MongoDB to create Similar Product Recommendations

    - Utilized OpenAI GPT3.5, Word2Vec, and Vertex AI to develop Bought Together Recommendations

    - Orchestrated ML Pipelines using Kubeflow and GCP Vertex AI for automated MLOps workflows

    - Deployed Recommendation ML Pipelines for freshpikonline.com and gofynd.com

Deep Learning Engineer

Paralaxiom Technologies Private Limited
Apr, 2022 - Nov, 2022 7 months

Associate Software Engineer

Paralaxiom Technologies Private Limited
Mar, 2020 - Mar, 20222 yr

Intern

Paralaxiom Technologies Private Limited
Dec, 2019 - Mar, 2020 3 months

    Pepsico Perfect Pallet & Pepsico Label Right Projects:

    - Trained Yolov5 PyTorch object detection Docker models across GPUs with MAP 96% on test set

    - Designed and implemented Mean Average Precision and under-sampling algorithm Python scripts

    - Maintained Model Registry for deployment using Azure CLI and Azure Blob Storage

    - Integrated Google Cloud Vision API for OCR and added custom ML logic to handle edge cases

Analyst

Arcesium
Jul, 2015 - Oct, 2015 3 months

Analyst

D. E. Shaw India Private Limited
Apr, 2015 - Jun, 2015 2 months

    - Reconciled daily and month-end Cash and Positions for D. E. Shaw MACRO trading desk

    - Assisted traders with ad hoc information requests regarding portfolio exposure, roll, value dates, and P&L impact

    - Collaborated with Pricing team to provide daily P&L estimates/deviations to the Traders and the Reporting team

    - Collaborated with Brokers, Traders, Middle Office, and the OTC team to resolve settlement issues for FX/FI products

    - Provided account allocations to Executing Brokers for MACRO strategy trades

    - Investigated and rectify concerns related to MACRO trade bookings and broker-side trade file information

    - Collaborated with IT to ensure smooth, efficient and fast reconciliation process

    - Conceptualized and executed a documentation project highlighting the major MACRO Fin Ops operational workflows


    Please Note: Arcesium was spun out of D. E. Shaw on July 1st, 2015

Education

  • Masters, Mathematics

    The University of Kansas, Lawrence, USA (2019)
  • Post Graduate Diploma in Management Finance

    IMT Ghaziabad, India (2015)
  • Bachelors, Mathematics and Physics

    The University of Kansas, Lawrence, USA (2012)