Shourya Sharma

Built an end-to-end RAG chatbot to increase student engagement up to 35%. The Chatbot ingests code exercises from 10000+ students, generating embeddings and storing them in MongoDB Atlas Vector Search to enable students to query code reviews, scores, and feedback in natural language. Designed a polymorphic provider architecture supporting interchangeable modules with 3 embedding models and 2 LLM backends (Ollama local / OpenAI cloud) with automatic fallback, which reduces inference cost to $0 via local GPU inference with Qwen 2.5 Engineered a fault-tolerant LMS scraper with auto re-authentication on token expiry ( 58 req), state persistence for resumable runs, and base64 payload decoding.

Automating data collection across all submissions with 0 manual intervention.

Gen AI engineer - Litcoder India - Remote

(2026-01)

• Built an end-to-end RAG chatbot to increase student engagement up to 35%. The Chatbot ingests code exercises from 10000+ students, generating embeddings and storing them in MongoDB Atlas Vector Search to enable students to query code reviews, scores, and feedback in natural language.
• Designed a polymorphic provider architecture supporting interchangeable modules with 3 embedding models and 2 LLM backends (Ollama local / OpenAI cloud) with automatic fallback, which reduces inference cost to $0 via local GPU inference with Qwen 2.5
• Engineered a fault-tolerant LMS scraper with auto re-authentication on token expiry ( 58 req), state persistence for resumable runs, and base64 payload decoding. Automating data collection across all submissions with 0 manual intervention.

Research Assistant - University Of Toronto - Toronto, ON – Remote

(2022-10)

• Spearheading the field of Generative A.I. Governance by developing detection systems for LLM jailbreak attacks, including crescendo, many-shot, and multi-turn manipulation techniques, achieving measurable reductions in Attack Success Rates.
• Designed a qualitative evaluation framework to measure Attack Compliance, utilizing semantic similarity detection. Used F-1 scores to quantify enforcement efficacy against adversarial threats. Stored the results in JSON files for ease of access and management.
• Implemented safety guardrails for RAG pipelines, establishing checkpoints at retrieval and generation phases to detect and mitigate prompt injection attacks.
• Collaborated with cross-functional teams to develop the ETAL website, utilizing SQL and JavaScript to manage data securely on University Library servers.

Data Scientist - University Of Toronto - Toronto, ON

(2022-10 - 2024-05)

• Researched and published the cognitive effects of VR simulations built in Unity and C# on 80+ participants, leading end-to-end data collection, thematization, and qualitative analysis using NVivo.
• Led data mining and management in Microsoft SQL Server with 1000+ queries across video and audio datasets. Accelerated audio transcription and speaker diarization by 12% using a Python-based NLP model built with PyTorch and Scikit-learn.
• Analyzed biostatistical time series data from 80+ VR participants using Pandas, Scipy, and Numpy for predictive modeling, and visualized findings using Matplotlib and Power BI.

Specialization Certificate - Agentic System and Design - Indian Institute of Technology Roorkee (2026-01)

B.Sc - Computer Science and Mathematics - University of Toronto (2017-09 - 2022-11)