shreyansh singh

Building agentic workflows with LangChain and LangGraph for real-world AI applications. Designing evaluation pipelines for LLM reliability, latency, and response quality, and refining prompt engineering and system design patterns.

Built a secure Python-based tax and accounting platform with compliance features and memory-safe computation. Integrated Azure AD with SAML SSO and designed LangChain/LangGraph agentic AI workflows using GPT-4o, managing scalable AWS CDK microservices on ECR/ECS.

Annotated and curated datasets focused on robustness, multilingual understanding, and edge-case handling. Designed prompt variations for generalization testing and identified weaknesses in long-context reasoning to improve response coherence.

Engineered high-performance banking API microservices with FastAPI and JWT auth to reduce transaction latency. Built OCR and face-matching computer vision pipelines, automated AWS operations with Boto3, deployed services via AWS CDK and ECR/ECS, and practiced TDD to reduce production defects.

Evaluated enterprise LLM outputs for productivity, structured data handling, and API use cases. Tested real-world scenarios such as documentation generation and debugging, and refined prompts to improve clarity and determinism.

Participated in RLHF evaluation workflows focused on alignment, safety, and instruction-following. Assessed outputs for harmful, biased, or unsafe content and refined evaluation benchmarks for technical and reasoning tasks.

Performed comparative evaluation and ranking of LLM outputs to improve factual grounding and reduce hallucinations. Tested model behavior with adversarial and edge-case prompts to improve determinism and consistency.

Created supervised fine-tuning (SFT) datasets with prompts, ideal responses, and reasoning chains for technical and analytical tasks. Standardized prompt formats to improve consistency, reproducibility, and generalization.

Evaluated LLM outputs for coding, debugging, and system-design tasks for correctness, reasoning depth, and edge-case handling. Designed structured evaluation rubrics and contributed to RLHF pipelines by ranking model responses.

Deployed end-to-end ML models using Python and FastAPI to accelerate deployment speed. Built an XGBoost-based heart-failure risk prediction model and automated cloud infrastructure with Terraform, adopting TDD to improve reliability and maintainability.

Earned a B.Tech in Computer Science Engineering from JSS Academy of Technical Education, Noida, graduating in July 2021.