SDE – AI/ML - Shopup (Fintech) - Bengaluru, India
(2025-11)
- Architected a Generator-Critic multi-agent system using Python, Agno (LLM orchestration framework), and FastAPI for financial invoice parsing across a BFSI platform, achieving 97% accuracy on 500K+ transactions/month with 40% false positive reduction via custom model evaluation checkpoints.
- Engineered a credential injection middleware for responsible AI security, preventing API keys and sensitive tokens from reaching LLM context windows; adopted across all internal AI workflows including PR review and coding assistants.
- Evaluated and deployed Temporal for real-time ML workflow orchestration over Airflow and Prefect, achieving sub-50ms latency and 30% higher throughput for production inference pipelines.
- Led architecture across 5+ GCP services; built SilqClaw — an internal AI agent orchestration platform with tool assignment, RBAC access control, and skill management for enterprise AI governance.
- Developed multilingual financial document extraction (English, Arabic, Bengali) for bank statements using FastAPI and NLTK, improving regional fintech adoption by 25%.
- Refactored AI agent infrastructure with OOP design patterns, reducing codebase complexity 30% across 18+ modules; mentored interns and conducted technical interviews (system design, algorithms).
AI Engineer (Founding Engineer) - Knowledge Keeper - Remote
(2024-08 - 2025-11)
- Built company AI infrastructure from scratch: RAG ingestion pipelines, inference services, document retrieval and semantic search systems, and REST APIs for enterprise document automation.
- Reduced VLM inference latency from 2.4s to 1.3s on GCP Vertex AI by migrating to vLLM with FP8 quantization, achieving 40% throughput improvement for production-scale inferencing.
- Built a multi-agent document automation system using CrewAI, Qdrant (vector database), and FastAPI with RAG architecture and embedding techniques, handling 10+ document types; reduced manual processing 80%.
- Developed real-time compliance monitoring pipeline using YOLO11 and Qwen3 VLMs on Kafka event-driven streams, replacing 75% of manual review workflows with automated AI evaluation.
- Fine-tuned DeBERTa, BERT, BART, Qwen2 for NLP tasks (sentence similarity, text generation) using custom model evaluation frameworks; achieved 97% production accuracy.
SDE – AI/ML - Shopup (Fintech) - Bengaluru, India
(2023-06 - 2024-08)
- Built ML classification platform for warehouse item categorization using FastAPI and Docker (containerized microservices), achieving 99% accuracy and 70% reduction in manual validation for the fintech supply chain.
- Designed OCR extraction APIs (PyTesseract, EasyOCR) for bank transaction slip processing in a BFSI context, reducing financial document processing time by 40%.
- Set up GitHub Actions CI/CD pipelines for automated testing and deployment of ML services, reducing release cycle time by 30% and enabling continuous model delivery (MLOps).
- Developed expense management and multi-level approval workflows (TypeScript, NestJS) used by 1,000+ operations staff daily on the CRT fintech platform.
SDE Intern - Shopup (Fintech) - Bengaluru, India
(2022-07 - 2023-06)
- Built Attendance Management System (Golang, AWS Rekognition) with facial recognition, reducing employee check-in time by 75%.
- Developed Risk Management system (Python, gRPC, Markov Models) for address validation in fintech operations, improving prediction accuracy by 75%.
ML Intern - AMTDC, IIT Madras - Chennai, India
(2021-12 - 2022-02)
- Trained object detection models (TensorFlow, Keras) across 10 retraining cycles with iterative model evaluation, reducing validation loss by 22%.
