B.S. Computer Science & Data Analytics, IIT Patna · Aspiring AI/ML Engineer · Full-Stack Developer
Building Intelligent Systems, Scalable Software & Real-World AI Products
Explore My Projects · Connect With Me
| Who I am | Computer science student at IIT Patna (B.S., 2029) — building AI products and reproducing ML research with verified numbers. |
| What I build | Fraud-risk call analysis, QR-based safety systems, ML reproductions, and civic-tech prototypes. |
| Strongest work | LoRA reproduction (full fine-tuning 0.9106 → LoRA r=16 0.8968 on 1.32% of parameters) · CallSuraksha AI (live) |
| Coding practice | 317 questions solved across LeetCode, CodeChef and HackerRank · CodeChef 1506 (2★) · LeetCode 88 (17 Hard) |
| Contact | sujalgiri5@gmail.com · LinkedIn |
I study computer science at IIT Patna while building practical AI software — systems that solve a real problem for a real user: a caller who might lose money to a scam, a venue that needs secure deposit handling, a city that needs to know which drain will flood first.
My work sits at the intersection of AI, machine learning, data science, and scalable web apps. I reproduce published methods and report the full result — including the seed-to-seed gap — and I do not claim a deployment, a result, or a role I cannot show.
Open to AI/ML and data science internships, software engineering internships, and research collaboration — including international and remote roles.
LoRA Reproduction — DistilBERT + SST-2
Reproduced LoRA with rank ablation r=4/8/16 against full fine-tuning. Full fine-tuning: 0.9106 accuracy. LoRA r=16: 0.8968 on 1.32% of parameters. All-linear r=16: 0.9071 — only 0.35pp below full fine-tuning. Seed gap reported honestly as 1.4–2.4pp. 9 runs, fully reproducible.
Python · PyTorch · Hugging Face · PEFT
CallSuraksha AI — "Call Suno, Risk Samjho." · Live demo
AI-assisted scam-risk analysis for recorded Hindi / Hinglish / English calls: transcription, explainable risk flags (Low / Medium / High / Unable to Assess), and spoken Hindi safety guidance. A risk flag — never a fraud verdict. No live-call interception, no deepfake detection, no caller identification.
React 19 · TypeScript · Express · Gnani Prisma (speech-to-text) · Gnani Timbre (voice) · rule-based risk engine
SafeTag — QR-Based Secure Deposit System · Landing page
QR-based deposit handling with WhatsApp alerts for temples, exam centres, parks and museums — built with Team HackShield. Status, honestly: a controlled-pilot candidate — production launch is not approved and no deployment is authorised yet. The live link is the landing page, not a running product.
TypeScript · Next.js · MongoDB
NalaSetu — Predictive Municipal Drain Cleaning
A prototype that prioritises which drains to clean before heavy rain: risk score from rainfall forecast, blockage history, terrain and citizen reports, with a Leaflet map and crew-dispatch workflow. Prototype with synthetic demo data — not operational municipal use.
React 19 · TypeScript · TanStack Start · AWS Lambda / DynamoDB (code)
- CampusMate AI — AI study companion (topic explainer, study planner, practice questions) with Google Gemma. Next.js · TypeScript · Tailwind
- RideBuddy — Student ride-sharing platform with JWT auth. React Native · Node.js · Express · MongoDB
- FlowMind AI — AI workflow automation with Gemini, PostgreSQL and n8n (single-commit showcase; demo mode uses sample data). TypeScript · Next.js
- Customer Churn Prediction — Churn prediction dashboard with ML and business insights. Python · Streamlit · scikit-learn
- CodeVerse — 3D computer-science learning game, in development (private repo, not linked). Three.js · React Three Fiber
Verified from Codolio — SujalIITP (11 Nov 2026): 317 questions solved across platforms — LeetCode 88 (Easy 32 · Medium 39 · Hard 17, rating 1500, max 1720), CodeChef 176 (rating 1506, 2★ · DSA rating 1817), HackerRank 44 (all C++). 522 submissions · 77 active days · max streak 35 · 17 contests. Codeforces (9 solved) and GeeksforGeeks (0) are thin — the gap, left visible.
Only technologies demonstrated in actual projects. Currently exploring: Quantum Computing · LLM / RAG systems · Efficient fine-tuning.
Public account, verified 11 Nov 2026: 51 public repositories, 0 stars, 3 followers. No stars claimed or hidden — the work above is the signal.
- 3rd Place (91 marks) — Innohack 3.0 for QubitCraft — quantum computing learning platform, Team SRIJAN (Lead Developer). Its repository belongs to a teammate; linked as an achievement only.
- SIH 2026 — Team SRIJAN, Problem Statement SIH26140 (QubitCraft)
- LoRA reproduction — 9 runs, fully reproducible, results verified run-by-run
- B.S. Computer Science & Data Analytics — IIT Patna (2029) · CPI 9.18/10
- B.Tech Computer Science & Engineering — SIET (AKTU) (2028) · concurrent degree
Open to research collaborations, hackathons, and internships — if you're building something in ML, quantum, or AI for Bharat, let's talk: sujalgiri5@gmail.com