class KinshukGupta:
institute = "IIT Bhilai — BTech, Data Science & AI (2023–2027)"
cgpa = 8.48
location = "Bhilai, Chhattisgarh 🇮🇳"
currently_building = [
"Vision42 — Vision-Language System for Satellite Imagery (ISRO SAC)",
"CauESC — RL + Persona-Attention for Emotional Support Dialogue",
]
open_to = ["CV Research Collabs", "YOLO-OBB / Rotated Object Detection", "Open Source"]
obsessions = ["Algorithmic Efficiency", "Multimodal AI", "Big Data Pipelines"]
fun_fact = "2000+ DSA problems solved. Still not enough. 💀"| 🥇 Achievement | 🔢 Result |
|---|---|
| JEE Advanced | AIR 6309 |
| JEE Main | 99.20 Percentile |
| Inter IIT Tech Meet 14.0 — ISRO SAC GeoNLI | 🥈 6th Rank Nationally |
| Inter IIT Tech Meet 13.0 — ISRO Lunar Mapping | 🏅 11th Rank Nationally |
| IICPC 2025 — Shaastra (IIT Madras) | Rank 1102 + Finalist |
| The Forge Hackathon — OpenLake × GDG Meraz 6.0 | 🥉 3rd Position |
| DSA Problems Solved | 2000+ across platforms |
🛰️ Vision42 — Satellite Vision-Language System (ISRO SAC × Inter IIT 14.0)
Stack:
PythonFastAPIOpenCVYOLO-OBBVLMsMultimodal Fusion
- Architected a vision–language system for satellite imagery supporting image captioning, VQA, and text-driven object grounding
- Designed a hybrid OBB + multimodal fusion pipeline handling rotated objects and scale variability
- Evaluated on IoU and BLEU-3 metrics — Secured 6th Rank nationally
🌕 Lunar Surface Mapping — Chandrayaan-2 XRF Data (ISRO × Inter IIT 13.0)
Stack:
PythonNumPySciPyAstropyPyTorchQGIS
- Engineered a pipeline to process Chandrayaan-2 CLASS XRF data → high-resolution lunar elemental composition maps
- Custom clustering + spatial filtering to resolve noise and overlapping scan regions
- Accelerated with PyTorch parallelization — Secured 11th Rank nationally
💬 CauESC — RL + Persona-Attention for Emotional Support Dialogue
Stack:
PythonPyTorchTransformersReinforcement LearningNLP
- Enhanced emotional support dialogue with COMET-based commonsense reasoning + persona conditioning
- Designed Persona Attention Loop (PAL) — dual cross-attention decoder for strategy-aware response generation
- Two-stage training: MLE (pointer-generator) + REINFORCE RL
- Achieved BLEU-4: 3.92, ROUGE-L: 16.91 on ESConv dataset
📚 StudySphere — Focus Efficiency Hackathon App (3rd Place, Meraz 6.0)
Stack:
ReactViteJavaScript
- Built a productivity application featuring a custom 'Focus Efficiency' algorithm to quantify user behaviour
- Won 3rd Position at The Forge Hackathon (OpenLake × GDG)
AI / ML / DL
Languages
Big Data & Cloud
Tools & Infra
| Platform | Handle | Highlight |
|---|---|---|
| 🟡 LeetCode | kinshuk18 | 500+ problems solved |
| 🔵 Codeforces | kinshuk18 | 300+ problems solved |
| 🟠 CodeChef | kinshuk18 | 3* Max Level |
| 🟢 GFG | kinshukgucn44 | 500+ problems solved |
📡 Distributed Systems ████████████░░░░ Apache Spark + Kafka at scale
🛰️ Vision-Language Models ███████████░░░░░ Beyond CLIP — custom VLMs
🤖 RL for NLP ████████░░░░░░░░ Production-grade REINFORCE loops