CV

Education

University of California, Riverside — MS in Computer Science
September 2025 – December 2026 · Riverside, California

Indian Institute of Technology, Bombay — B.Tech in Computer Science
August 2019 – April 2023 · Mumbai, India

Experience

Research Intern: CompFly AI

June 2026 – September 2026 · San Francisco, California

• Built a reproducible benchmark harness for governance controls across AgentDojo prompt-injection attacks and ToolEmu high-stakes tool-use scenarios.
• Benchmarked and tuned a lightweight transformer-based prompt-injection detection pipeline, including chunking, window/stride settings, heuristic rules, and parallel OR ensembles.
• Extended OpenTelemetry instrumentation across agent orchestration, LLM inference, and tool execution for trace-level analysis of agent turns, tool calls, handoffs, token usage, and cost.
• Contributed the technical write-up Evaluating Prompt-Injection Controls Beyond Detection to the CompFly blog.
• Explored governance for memory-enabled agents and collaborated on experiments for an evolving deepfake detection/generation project.

Data Scientist: Finarb AI

August 2023 – August 2025 · Kolkata, India

• Implemented and benchmarked SARIMA, VAR, Prophet, LSTM, and Temporal Convolutional Network models for pharmaceutical supply-chain demand forecasting, reducing mean absolute percentage error by 90% over existing moving-average and exponential-smoothing baselines.
• Developed an end-to-end forecasting pipeline covering preprocessing, feature engineering, custom-metric model selection, and Docker-based deployment across supply-chain nodes.
• Built AI agents for RAG-based feature engineering, natural-language cross-database queries, SQL/Python code generation, and automated KPI dashboards.

Selected project

AI for Navigation: Comparative Study of LLM Reasoning Strategies

Spring 2026 · GitHub project files

• Co-authored a controlled study comparing Reactive, ReAct, Chain-of-Thought, Tree-of-Thought, and Buffer-of-Thought strategies for sequential decision-making in MiniGrid navigation tasks.
• Built a deterministic Python evaluation pipeline around Qwen2.5-7B-Instruct, tracking success rate, exploration steps, and token usage over 20 episodes per environment.
• Implemented Buffer-of-Thought with structured outputs, reusable navigation templates, and safety checks for unsafe forward actions.
• Analyzed planning-performance trade-offs across task difficulty, reaching up to 55% success on LavaGapS6 and 20% on LavaCrossingS9N2.

Selected presentation

• LLM Safety & Mechanistic Interpretability — prepared and presented a graduate-level, paper-based lecture for EE 269 at UC Riverside with Professor Yinglun Zhu, covering four recent papers across LLM safety, mechanistic interpretability, and their intersection.

Selected open-source projects

• SkillFalsifier — built a falsification gate that tests whether self-generated agent skills transfer to held-out and adversarial cases before promotion.
• Terminal DreamGym — developed a local training loop that turns terminal-agent failures into practice worlds, recovery skills, and transfer evaluations.
• Scalable Analysis of Vegetation Anomalies — contributed to a cloud-native Big Data pipeline using Sentinel-2/Landsat imagery, AWS S3, PostGIS, PySpark, and a Flask/Leaflet dashboard; report.
• Image-realiser — built a CLIP-based discriminator and convolutional purifier with adversarial training, residual maps, and Grad-CAM explanations for synthetic-image forensics.
• Institutional memory in LLM systems and LM-JEPA symbolic regression — explored governance and path dependence in multi-agent memory, and structured generation for scientific formula discovery.

Teaching

Reader/Grader: CS 228: Deep Learning

Spring 2026 · University of California, Riverside · Professor Zhouxing Shi

As a graduate student at UC Riverside, I evaluated assignments and exams, provided feedback, and contributed rubric ideas. Relevant coursework also includes CS 229: Advanced Machine Learning.

Technical skills

• Languages: Python, C++, R, JavaScript, Bash, Prolog, AWK, MIPS, Neo4j CQL, SQL, sed, Lex/Yacc, LaTeX
• ML and data: PyTorch, TensorFlow, Pandas, NumPy, SciPy, MATLAB, Power BI, PostgreSQL, MongoDB
• Software and platforms: React, Django, Docker, Git, Microsoft Excel, RStudio, DataGrip, PowerPoint, Word, Slack, Azure DevOps, Jira, Confluence

Accomplishments

• Ranked first in OBC-PwD at IIT-JEE Advanced 2019 and placed among the top 0.8% nationwide in JEE Main among 1.3 million candidates.