R

Rishi

About

Detail

Jharkhand, India
Contact Rishi regarding: 
id_card
Internships

Timeline

work
Job
school
Education
folder
Project
flag
Award

Résumé

Jobs verified_user 0% verified
  • G
    Open Source Developer
    Godspeed Systems
    Jan 2025 - Feb 2025 (2 months)
    Developed a CLI tool for LLM-assisted debugging with full codebase context and accessible TUI for precise issue resolution. Implemented TreeSitter based AST parsing to extract function/class definitions for the entire codebase for code mapping. Engineered repository mapping via PageRank algorithm on AST to rank relevant code sections for token-efficient context. Implemented complete state management with persistent conversation history and multi-turn conversation support.
Education verified_user 0% verified
  • Indian Institute of Technology, Kanpur
    B.Tech
    Indian Institute of Technology, Kanpur
    Jan 2023 - Current (3 years 10 months)
    CPI/10: 7.2/10
  • D
    CBSE (XII)
    Delhi Public School, B.S.City
    Jan 2022 - Dec 2022 (1 year)
    97%
  • D
    CBSE (X)
    Delhi Public School, B.S.City
    Jan 2020 - Dec 2020 (1 year)
    96.2%
Projects (professional or personal) verified_user 0% verified
  • J
    JokeSearch
    Jul 2025 - Sep 2025 (3 months)
    Tree based LLM state space exploration for novel joke generation.
  • KGEN
    Web3 Wallet Persona Analysis
    KGEN
    Jun 2025 - Jul 2025 (2 months)
  • G
    Game Hacking
    May 2025 - Jul 2025 (3 months)
    Mentored 20+ students in computer architecture, binary exploitation, and reverse engineering fundamentals. Demonstrated decision tree models to reverse-engineer game levels enabling targeted brute-force with reduced search space. Demonstrated network traffic interception and packet manipulation techniques to modify game logic in a live game environment
  • L
    LLM Generated Text Detection
    Mar 2025
    Explore methods to reliably distinguish LLM-generated text, robustly with a low False Positive Rate(FPR). Developed DeBERTa-v3 detector using Contrastive Learning on RAID dataset to learn style embeddings. Utilized triplet-based dataset preparation with anchor-positive-negative pairs using content/topic matching. Implemented training-free few-shot adaptation using style centroids for boosted accuracy in OOD regimen. Achieved 96% Accuracy with
  • P
    PNB Cybersecurity Hackathon 2024
    Dec 2024 - Feb 2025 (3 months)
    Built local high-performance cross-platform antivirus to protect public sector banks from ransomware. Implemented a real-time detection engine with hash matching and YARA rules for pattern-based detection. Utilised SQLite3 for efficient local storage of 1M+ malware hashes, reducing application memory usage. Automated threat intelligence sync via MalwareBazaar API, adding newer hashes to threat database. Implemented whitelist filtering and pre-compilation of YARA rules to optimize application performance. Analyzed 1M+ malware samples from the threat database to generate interactive trend visualizations. Achieved 1st place nationally, winning INR 5,00,000 plus Global Fintech Forum (GFF'25) invitation.
  • D
    Dynamic Agentic RAG with Pathway
    Oct 2024 - Dec 2024 (3 months)
    Developed a comprehensive, efficient, and robust end-to-end dynamic Agentic RAG pipeline. Implemented a Planner Agent to dynamically create and orchestrate sub-agents in a quasi-parallel DAG. Integrated financial/legal sources via streaming document store for live-indexing and agentic RAG on chunks. Developed a RAG pipeline with hybrid search, fusing BM25/HNSW via RRF and dedicated reranker. Generated reports with 35-40k tokens, $0.007/report, leading to IIT Kanpur's 6th position among 23 IITs.
  • L
    LinkedinBot
    Oct 2024
    Integrated Web Search, Web Scraping, and LinkedIn Posting functionalities for autonomus LinkedIn content creation. Leveraged LangGraph and the function calling features of GPT-40-min for recursive tool invocation and agent orchestration. Experimented with Llama 3.1, Command-r models and Groq/Fireworks inference services to optimize per-turn latency.
  • O
    OptiParse
    Aug 2024 - Sep 2024 (2 months)
    Built receipt processing pipeline using PandaOCR for extraction and LLM-based few-shot prompting for JSON structuring. Created synthetic ground-truth dataset using high-accuracy Multimodal LLMs for annotating existing receipt images. Measured performance via Levenshtein distance against ground-truth, deployed on Azure App Service/AI for inference.
  • N
    NFS Goes Online
    May 2024 - Jul 2024 (3 months)
    Built a persistent file system from scratch in C, implementing core on-disk structures including an inode table, data blocks etc. Implemented a client-server architecture over UDP sockets to handle remote file operations (create, read, write etc.).
Awards verified_user 0% verified
  • 1
    1st place nationally
    Feb 2025