Sanyam Wadhwa
Software Engineer
Education
Thapar Institute of Engineering and TechnologyAug 2023 to 2027 (expected)
B.E. Computer Engineering, CGPA 8.31Patiala, Punjab
Experience
Software Engineering InternFeb 2025 to Nov 2025
TIET Examination Platform, exams.thapar.eduPatiala, part time
- Built the seating and invigilation workflow of TIET's 5-developer Flask exam platform that seats all 12,000+ students for every paper in a cycle, cutting each plan from 3 to 4 days of manual work to about 40 minutes.
- Rewrote 2 of the platform's 4 seating algorithms, AB and ABCD, so when a course runs out mid-room, waiting courses take its seats in order of students left, and rooms down to 2 courses switch to a diagonal 2x2 layout.
- Shipped the exam office's change tools to block unavailable rooms, swap rooms and invigilators and print per-teacher duty PDFs, and moved their inputs from browser sessions into MySQL so all admins share one saved state.
AI Research InternJun 2025 to Jul 2025
Centre of Excellence - UQ, TIETPatiala, on site
- Designed a mental-health chatbot's safety path as plain rules instead of a model, with a keyword crisis gate that returns helpline numbers before any LLM call and override rules that route every other message.
- Built the message pipeline that runs fine-tuned BERT emotion and sentiment classifiers and several mental-health tests in parallel, then has the LLaMA responder write a report from their combined results.
- Served the responder as a 4-bit LLaMA via llama.cpp, so the prototype ran locally without a paid API.
Projects
- Built a live coding-interview platform where interviewers and candidates code in a shared editor, on an operational-transformation sync engine written from scratch and checked by 30,000 randomized tests per push.
- Cut lost keystrokes from 49% to zero for two people typing at once on the live site by merging their edits instead of resending the whole file.
- Logged every edit to PostgreSQL before confirming it, so rooms survive a server crash, dropped connections resume with no edit lost or doubled, and interviewers can replay any session.
- Built a research-paper reader that turns an arXiv ID into the paper's key findings with their reported numbers, plus its methods, datasets and figures, reading the full text instead of only the abstract.
- Achieved figure-extraction F1 of 0.972 on 16 arXiv papers, recovering all 122 vector and raster figures by growing each crop upward from its caption with PyMuPDF.
- Kept paper runs alive through API outages by failing over from Gemini to Groq to Ollama; a paper too large for the fallback's free-tier limit is reprocessed section by section instead of failing.
Technical skills
Languages: C, C++, JavaScript, Python, SQL, TypeScript
Backend and web: FastAPI, Flask, Node.js, SQLAlchemy, REST APIs, WebSockets, React, Next.js, Tailwind CSS
Data and infra: Docker, GitHub Actions, Linux, MySQL, PostgreSQL
AI: Hugging Face, LangGraph, llama.cpp, pgvector, PyTorch, RAG
Achievements and leadership
1st place, Clinical/Engineering track, PGIMER and AOPN Advances in Pediatric Epilepsy, for NeoSpasmApr 2026
Winner, Samsung PRISM Web-Agent Hackathon, for KONTA, on-device tab search built by a team of 4Mar 2026
Top 50 of 1,000+ on Eris, shipd.ai's Kaggle-style ML platform, with 10 gold, 8 silver and 9 bronze medalsOngoing
Top 200 in 2025 and Top 300 in 2026, Amazon ML Challenge2025, 2026
Technical Coordinator, Microsoft Learn Student Chapter, TIETAug 2023 to Apr 2025