Ishan Tiwari
Researching reinforcement learning and documenting the insights. Exploring and validating applied AI workflow solutions that bridge experimental agents and production-grade systems.
About
I'm a recent Computer Science graduate from the University of St. Andrews and researcher passionate about the intersection of API development, AI systems, and developer productivity. My work focuses on creating intelligent tools that streamline API integration workflows and enhance the developer experience through advanced automation and AI-powered solutions.
My dissertation research (initiated May 2024) focused on solving the persistent challenge of inconsistent API documentation and the manual complexity developers face when integrating multiple APIs into their applications. I developed an AI-assisted framework for automating API integration using Retrieval-Augmented Generation (RAG) with CodeLlama-13B-Instruct, combining real-time API specifications retrieval with contextual code generation. The system achieved an 86.9% test pass rate across 315 comprehensive tests, dramatically reducing integration time while maintaining code quality and security standards.
Key Achievements: Advanced through PRMO to Regional Mathematical Olympiad (RMO) in India, ranking among top 300 students across two states, UKMT Senior Kangaroo Mathematics Challenge Gold Award (2020/21), selected among top 100 students nationwide for prestigious "Raising a Mathematician" residential camp in India, won the Dedalus Labs track at the AI Tinkerers Hackathon (supported by Anthropic, ElevenLabs, and Vibe Kanban) with MCP Games, a Gymnasium-to-MCP tooling concept for unified agent training workflows, football team captain leading squad to citywide and regional runners-up position, and hosted The Student Box podcast with episodes spanning diverse academic topics from modern science to classical civilisation.
In my spare time, I previously hosted The Student Box podcast, exploring topics from modern science to classical civilization, and contribute to projects focused on AI and developer tools.
Research & Projects
MCP Games - AI Tinkerers Hackathon Winner
Built MCP Games during the AI Tinkerers Hackathon, winning the Dedalus Labs track supported by Anthropic, ElevenLabs, and Vibe Kanban. The platform converts Gymnasium reinforcement-learning environments into Model Context Protocol tools so AI agents can train and interact through a unified interface, proving out shared telemetry, curriculum design, and evaluation loops. Explore the code on GitHub.
Model Context Protocol
Reinforcement Learning
Gymnasium
AI Agents
Bibliography Database Transformation
Research internship with University of St. Andrews and Griffith Institute, Oxford. Automated digitization of bibliographic references for Egyptian hieroglyphic texts using enhanced AnyStyle parser with integrated validation mechanisms. This work aimed to support the academic community by providing improved access to a database of Egyptian hieroglyphic texts.
Python
Data Processing
Parser Enhancement
Academic Research
Fancy-tree
Git-aware tree command that goes beyond file structure by parsing and displaying functions, classes, and code structure inside files. 2k+ downloads with cross-language code analysis powered by tree-sitter. Supports 160+ programming languages and respects .gitignore files, making it the perfect tool for giving AI systems comprehensive codebase context without manual explanation.
Python
Tree-sitter
Multi-language Parsing
Git Integration
PyPI Package
Young Enterprise: NFC Wi-Fi Connectivity Solution
Led technical development and sales strategy for custom-designed NFC-enabled Wi-Fi connectivity device in Young Enterprise national competition. Sold out entire product inventory (200+ units within 3 months) and advanced to regional finals winning the best overall team, competing against hundreds of student ventures across the UK. Won multiple awards at trade fairs through innovative product demonstrations and strategic market positioning. Responsible for both technical architecture and go-to-market strategy, managing product development from concept to successful commercialisation.
NFC Technology
Hardware Design
Sales Strategy
Market Research
Product Management
AI-assisted Framework for Automating API Integration
Dissertation research developing an AI-assisted framework using Retrieval-Augmented Generation (RAG) with CodeLlama-13B-Instruct. Implemented web scraper for public APIs, a dual-database architecture (relational and vector stores) with comprehensive static analysis pipeline including AST-based evaluation. Achieved 86.9% test pass rate across 315 tests spanning five distinct API domains, demonstrating superior efficiency in automated code generation.
RAG
CodeLlama-13B
Vector Databases
AST Analysis
API Integration
NEAT Algorithm Implementation
Successfully implemented NeuroEvolution of Augmenting Topologies (NEAT) algorithm for evolving neural network architectures. Trained agents to consistently beat internal AI opponents in Neural Slime Volleyball, demonstrating emergent behaviors and network complexity evolution through genetic algorithms. Achieved stable convergence and documented the evolution of increasingly sophisticated network architectures over multiple generations.
Python
JAX
Neural Networks
Genetic Algorithms
Japanese Kanji Generation via Stable Diffusion
Fine-tuned Stable Diffusion models to generate novel Japanese Kanji characters from English definitions, successfully creating culturally meaningful characters for modern concepts like "Gundam" and "Deep Learning." Built comprehensive dataset pipeline converting thousands of SVG Kanji to optimized pixel format, implementing custom data validation and quality control measures. Achieved high-quality character generation that maintained traditional stroke patterns while representing contemporary concepts.
Stable Diffusion
PyTorch
Computer Vision
Data Engineering
CodeInsight: AI-Powered Developer Assessment Platform
Revolutionizes technical hiring by providing deep insights into how candidates actually code, solve problems, and collaborate with AI tools. Unlike traditional coding assessments that only evaluate final output, CodeInsight analyzes the entire development process to reveal true engineering capabilities. Features AI collaboration intelligence, development workflow analysis, code quality assessment, and advanced time analysis to distinguish between AI-assisted engineers and AI-dependent coders.
FastAPI
React + TypeScript
VS Code Extension
AI Analysis
Behavioral Analytics
Agentic Automated SDK Generation
Currently exploring agentic automated SDK generation to workflow orchestration systems. This project focuses on creating intelligent agents that can automatically generate SDKs from API specifications and orchestrate complex workflows. The system uses advanced AI techniques to understand API patterns, generate high-quality client libraries, and create seamless integration workflows for developers.
AI Agents
SDK Generation
Workflow Orchestration
API Integration
HackHub - Unified Hackathon Discovery Platform
Built as a post-evening hack within 2 hours and open-sourced, this unified hackathon project discovery platform aggregates projects from DevPost, GitHub, and various hackathon events into one browseable interface. Designed as a "GitHub for hackathon projects" with advanced filtering by tech stack, event, and category. Features GitHub repository analysis, project voting, and community-driven discovery to help developers find innovative projects and collaborate on them to build their hackathon portfolio.
Key Idea being that the projects don't remain stale and can be imporved upon by the community.
Web Scraping
GitHub API
Rapid Prototyping
Open Source
Error Context Collector
VS Code extension that automatically captures Python error context and integrates with Cursor AI for intelligent debugging. Features smart context collection, symbol analysis, and zero-configuration setup. Successfully published on VS Code Marketplace and Open VSX, serving the global developer community with production-ready debugging enhancement tools.
TypeScript
VS Code API
Python Analysis
Cursor AI
World of Wikidata Weather Visualization Tool
Developed an interactive weather data visualization platform using WikiData in collaboration with the Wikimedia Foundation team in a 21-member, multi-team software engineering project. Ensured product alignment with Foundation user stories and accessibility requirements, implementing WCAG compliance and inclusive design principles. Utilized advanced data visualization techniques with 3D plotting capabilities, Agile and Scrum methodologies for efficient project management and seamless software integration across multiple development teams.
Data Visualization
WikiData API
Accessibility (WCAG)
Agile/Scrum
Foundation Collaboration
Ventures
Proteus
2024
Founding Team Member
Was part of the founding team for a venture-backed startup attempt focused on no-code API monitoring solutions. Built a platform as a service for custom monitoring rules, regional performance tracking and API behavior analysis.
Go
JavaScript
API Design
No-Code Platforms
PanScience Innovations
2024
Full-Stack Developer
Led full-stack development of B2C Faith Tech project at AI-driven startup incubator. Responsible for complete technical execution across front-end and back-end systems, ensuring seamless user experience.
Full-Stack Development
AI Integration
B2C Platforms
ConceptVines
2023
Algorithm Developer
Designed advanced decision-making algorithms focused on reputation factors for product engineering initiatives. Created scalable solutions supporting high-quality software engineering for startups and enterprises.
Algorithm Design
Python
Machine Learning
Inferyx Inc.
2020
ML Engineer Intern
Built extra trees classifier for HR analytics at NYC startup focused on automated data and ML pipelines. Developed employee attrition forecasting models using statistical analysis and machine learning.
Machine Learning
Pandas
Statistical Analysis
Insights
Standardized RL Environments and Multi-Agent Research
2025
Exploring plug-and-play RL ecosystems so LLM-driven agents can train, evaluate, and hand off work without bespoke environment setup.
Building fancy-tree to eliminate the repetitive task of explaining codebase structure to AI tools every single time
2025
Why I chose IPFS to publish my previous website – On using IPFS as a decentralized alternative to the centralized web.
2023