Skip to content
View amkabir's full-sized avatar
  • EY

Highlights

  • Pro

Block or report amkabir

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
amkabir/README.md

Kabir Gupta

CS Honors student · GenAI engineer · Legal-IR researcher

I build retrieval and LLM systems — RAG pipelines, hybrid search, rerankers, agents — and I like knowing what's underneath them, down to fork-per-connection servers and inode tables in C.

📍 B.Sc. Computer Science Honors @ University of Alberta (Dean's List, class of 2027), with an exchange year at Cal Poly San Luis Obispo.

Now

  • 🏗️ Generative AI Engineer Intern @ EY — building an LLM-based worker training & assessment agent for industrial manufacturing: a RAG pipeline over SOP and safety documentation (Python, FastAPI) that generates role-specific assessments, grades free-text answers, and surfaces skill gaps to supervisor dashboards.
  • 🔬 Research Intern @ Hokkaido University, Knowledge Base Lab — legal case retrieval for the COLIEE benchmark (details below).
  • ⚖️ Founder @ University of Alberta Startup Innovation Center — an on-prem, privacy-first AI platform for compliance-heavy legal workflows: structured case-memory extraction (spaCy NER, semantic deduplication, Pydantic validation) and hybrid document + graph retrieval. Shaped by 10+ interviews with law firms.

Research: legal information retrieval (COLIEE)

Designing and benchmarking multi-stage prior-case retrieval pipelines over 1,870 Canadian Federal Court decisions for the COLIEE legal case retrieval task:

  • Lexical stage — BM25 + TF-IDF with legal-aware tokenization, CombSUM fusion, and learned cutoff prediction
  • Dense + hybrid stage — BGE embeddings with FAISS indexing fused with the lexical signals
  • Reranking & judging — cross-encoder reranking, Random Forest candidate classification, and Qwen-based LLM relevance judging for final precedent selection

Code is private for now under the lab's data-use agreements — happy to talk about it.

Featured projects

Project What it is
AutoPhisher Phishing-simulation & security-awareness training platform (React + Flask). Multi-LLM email generation via OpenRouter with strict training-only guardrails. Built with a Cal Poly team.
langgraph-quiz-app Quiz platform on LangGraph agent graphs + Streamlit + SQLAlchemy, running fully local on Ollama — no cloud API keys.
Event Lottery System Android app for fair, lottery-based event registration — Firebase backend, QR check-in, organizer/entrant/admin roles. Team project, agile with CI.
multiprocess-http-server HTTP server in C on raw POSIX sockets with a fork-per-connection process model. No frameworks.
file-system-emulator Unix-style file system emulator in C — inodes, directories, and file operations over a virtual disk.

Previously

  • Software Developer Intern @ Interra Systems — multi-product RAG documentation assistant for media QC/monitoring platforms: 1,000+ pages indexed into per-product FAISS knowledge bases, conversation-aware query rewriting, CrossEncoder reranking, a 4-criteria answer-verification agent, and an enterprise FastAPI + Vue.js stack with analytics that surface documentation gaps.
  • Software Developer @ Ayodo Foundation — privacy-preserving "Face Type" soft authentication: an OpenCV pipeline (YuNet detection, 68-point landmark fitting, canonical alignment, normalized geometry features) classifying faces into 12 coarse geometric categories instead of identifying individuals.
  • Medley Interlisp (open source) — proposed benchmark additions (NEW-BENCH suite) to the Gabriel Lisp benchmarks used by the Medley Interlisp revival project.

Stack

Languages — Python · Java · C · C# · JavaScript · SQL · Lisp AI / retrieval — LangChain · LangGraph · FAISS · BGE embeddings · cross-encoder reranking · BM25 · OpenAI / OpenRouter APIs · scikit-learn Backend / frontend — FastAPI · Flask · React · Vue.js · Streamlit · SQLite · Firebase Systems — Unix/POSIX · sockets · processes · Make · Git

Contact

Email · LinkedIn · Toronto, ON

Pinned Loading

  1. amkabir amkabir Public

    Profile README