Skip to content
View Eishaan-Khatri's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report Eishaan-Khatri

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
Eishaan-Khatri/README.md

Eishaan Khatri — ML researcher & systems builder

Portfolio Blog Scholar LinkedIn Email

typing


🧭 About

I'm a final-year Mathematics & Computing student (Rank 1 in my department) who works in the space between a research paper and a production system — training models that hold up under distribution shift, and shipping the pipelines that carry them.

My work spans multilingual & code-mixed NLP, model robustness, edge AI, and multi-agent LLM systems. I care about honest metrics, reproducible experiments, and software that keeps working after the demo ends. Three of my projects have turned into peer-reviewed publications; many more live in the repos below.

role:        ML researcher & systems builder
education:   B.Tech Mathematics & Computing · RGIPT (Rank 1) · 2022–2026
focus:       robustness · evaluation · multilingual NLP · agentic systems
publications: 3  (ESWA · CASML · ICDM)
principle:   honest metrics  >  hero numbers
open to:     research-associate / applied-scientist roles · 2026
location:    Hapur, India  ·  open to relocation

💼 Experience

  • Proscend Communications IndiaEdge-AI & Video Analytics Intern (B.Tech Thesis) · Feb–May 2026 Built Edge-VCA, a CPU-only video-analytics pipeline (YOLOv8-Nano + a custom tracker) hitting 24.5 FPS at 185 MB RAM — replicating and extending a commercial surveillance system.
  • Polka Labs · Polkassembly (Singapore) — AI Developer Intern, Remote · May–Nov 2025 Shipped multi-platform governance & persona bots — multi-agent LLM voting, a 9-agent persona factory, Mem0 long-term memory — and designed a multimodal RAG memory architecture.
  • IIM ShillongResearch Analyst Intern, Remote · May–Dec 2025 Explainable-AI pipelines (SHAP / LIME / attention) over XGBoost & LightGBM, with Tableau dashboards and stakeholder-facing model interpretation.
  • Earlier AI/ML internships (2024): AspireNex, Lizmotors, Sambin — applied ML tasks, model building, and data pipelines.

🎓 Education

  • Rajiv Gandhi Institute of Petroleum Technology (RGIPT) — B.Tech, Mathematics & Computing + Minor in Business Consulting · 2022–2026 · Rank 1, Dept. of Mathematical Sciences.
  • Delhi Public School, Hapur — CBSE · Class XII 90.67%, Class X 96.80%.

📄 Publications

Year Venue Work
2026 Expert Systems with Applications · Elsevier Q1 Span-level target & harmfulness detection in code-mixed Hinglish — built the 6,800-post HINTOS dataset ⧉
2025 CASML · poster Enhancing & evaluating zero-cost LLMs for scientific QA — ~68% hallucination reduction ⧉
2025 ICDM · camera-ready ML-based network intrusion detection over large-scale IDS flow datasets ⧉

🧪 Featured projects

Project What it is Stack
IRMDS Real-time monitoring & decision system — kernel + 4 anomaly modules, REST/WS API, dashboard (98 tests, CI, Docker) FastAPI YOLOv8 Docker
PRISM / Verity-Ranker Evidence-backed hiring ranker — 100k candidates in seconds, deterministic FAISS BM25 RRF LLMs
Edge-VCA CPU-only edge video analytics — 24.5 FPS at 185 MB RAM, custom tracker (thesis) YOLOv8 OpenCV
IACV TTA Study Test-time adaptation on CIFAR-10-C — when adaptation helps vs. hurts PyTorch Tent
IACV CLIP Robustness Zero-shot CLIP robustness & calibration under shift OpenCLIP ViT
sql-migration-env Adversarial RL benchmark for LLM DB-migration agents (OpenEnv) RL SQLite Docker

➕ event extraction & cross-lingual transfer · recsys + off-policy eval · uncertainty estimation · industrial anomaly detection — see all repos.

🛠️ Skills & tools

Languages Python R C++ SQL TypeScript
ML / DL PyTorch TensorFlow scikit-learn HuggingFace
NLP / CV OpenCV XLM-R MuRIL Whisper CLIP YOLOv8 RAG
Systems FastAPI Docker Neo4j FAISS Streamlit GitHub Actions
Data Pandas Tableau Power BI XGBoost LightGBM stats & A/B testing

🏆 Achievements

  • 🥇 Rank 1 — Department of Mathematical Sciences, RGIPT
  • 🎓 MCM Scholarship (merit-cum-means)
  • 🧠 National Policy Hackathon 2024 — Finalist (IIM Calcutta & IIFT Kolkata)
  • 🌊 FloodAI Hackathon — Finalist (IIT Gandhinagar)
  • 📈 JEE Advanced 2022 (top ~1.5 percentile) · GATE 2026 (Data Science & AI) qualified
  • 3★ CodeChef · 🤝 Gyanarpan Teaching Volunteer (56 hrs)

📊 GitHub activity

GitHub stats Most used languages

contribution snake

🤝 Let's connect

I'm open to research-associate and applied-scientist roles (2026), research collaborations in robustness / multilingual NLP, and interesting problems in applied ML.

📮 eishaankhatri@gmail.com · 🌐 Portfolio · ✍️ Blog · 💼 LinkedIn · 🎓 Scholar

✦ Robustness over leaderboard chasing. Honest metrics over hero numbers. ✦  |  Beyond code: anime, lofi, and the occasional AI-made song. ✦

Pinned Loading

  1. sql-migration-env sql-migration-env Public

    A production-grade, adversarial evaluation suite for Database Engineering AI Agents. Features dynamic golden-database grading, live Mermaid ERDs, and 8 extreme SQLite migration tasks (OpenEnv).

    Python

  2. Cyclone-Preheater-Anomaly-Detection-System Cyclone-Preheater-Anomaly-Detection-System Public

    Industrial anomaly detection on cyclone-preheater sensor data — 378k rows, Isolation Forest, domain-informed features.

    Jupyter Notebook

  3. Reliance-Stock-Forecasting Reliance-Stock-Forecasting Public

    Gradient-boosted forecasting of Reliance / NIFTY-50 prices with directional-accuracy evaluation.

    Jupyter Notebook

  4. telegram-bot-v1 telegram-bot-v1 Public

    Forked from krishnaAiGen/telegram-bot-v1

    Python

  5. EishaanKhatri/drilling_mvp EishaanKhatri/drilling_mvp Public

    Python

  6. verity-ranker verity-ranker Public

    PRISM — evidence-backed ranking of 100k candidates against a job description: dense (FAISS) + sparse (BM25) retrieval fused with RRF and agentic scoring. Deterministic, CPU-only.

    Python 1 1