WORK_HISTORY
Production Engineering & AI Internships
Building scalable backend microservices, fine-tuning multilingual transformer models, and enforcing rigorous evaluation standards in real-world fintech, risk, and smart infrastructure platforms.
AI/ML Developer Intern
Hansraj Ventures · FinTech · LegalTech · Social Media
- Fine-tuned Llama 3.1 8B using LoRA with adapter merging for multilingual legal document translation, enabling automated regulatory compliance support across 8 Indian languages — Assamese, Bengali, Kannada, Gujarati, Nepali, Marathi, Tamil, and Malayalam — with measurably improved cross-lingual coherence.
- Architected the retrieval layer for Risk Minor, a multi-agent risk intelligence platform, combining BM25 keyword search with Milvus dense vector search behind a high-concurrency natural language query API — enabling sub-second regulatory document retrieval over large corpora.
- Implemented input/output validation guardrails and hallucination-detection fallbacks directly in the Risk Minor RAG pipeline, enforcing zero-tolerance for unsafe or fabricated regulatory context before any output reaches the client.
- Trained multivariate regression models for financial compliance and automated risk-scoring workflows; integrated them as FastAPI endpoints to eliminate manual analyst reporting steps.
- Built an automated RAGAS evaluation harness measuring faithfulness, answer relevancy, and context recall across all pipeline stages, with full Langfuse tracing for latency and quality observability across every inference node.
- Designed and deployed a computer-vision pipeline for cartoon-style image synthesis, optimized for constrained GPU VRAM execution using mixed-precision inference and model quantization.
- Engineered async FastAPI microservices with Docker CI/CD pipelines, achieving 40% reduction in request processing latency and sustaining 99%+ service uptime across all production endpoints.
⚡ 40% Latency Drop
🌐 8 Indian Languages
🎯 0.94 Faithfulness
🛡️ Zero-Hallucination Guardrails
📈 99%+ Uptime
Software Developer Intern
ACCESCO Living · FinTech & Smart Home Infrastructure
- Refactored a legacy monolithic codebase into async FastAPI microservices backed by Docker test environments, eliminating blocking I/O bottlenecks and reducing end-to-end API latency by 40% while improving unit test coverage significantly.
- Developed a regression-based inventory forecasting engine integrating real-time stock-level alerts — reduced stockouts by 35% and overstock carrying loss by 22% in simulated production scenarios.
- Built statistical anomaly detection pipelines over time-series IoT telemetry data using Pandas and Scikit-Learn, applying rolling z-score and IQR-based thresholding to decrease false-positive system alerts by 28%.
- Implemented reproducible ML training pipelines with TensorFlow and Docker volume mounts, ensuring full model traceability and version control across distributed datasets and training environments.
⚡ 40% API Speedup
📦 35% Stockout Reduction
📉 28% False-Positive Drop
🐳 Docker Testbeds
Technical Associate
SIT Photography Club
- Built and deployed the club's official web portal from scratch using modern HTML/CSS/JS, managing full lifecycle from design to deployment on a custom domain.
- Manage ongoing technical infrastructure including domain DNS configuration, SSL, and iterative content updates for club events and announcements.
Open-Source Contributor
GirlScript Summer of Code (GSSoC)
- Contributed across multiple open-source repositories — submitted bug fixes, performance optimizations, and improved documentation to developer tooling and machine learning utility libraries.
- Collaborated with maintainers via GitHub PR reviews, gaining experience with real OSS contribution workflows, code review culture, and async communication practices.