> python main.py
GAURAV
KASHYAP
Machine Learning EngineerOpen to ML and full-stack roles · India · remote-friendly
01 — About
About Me
I build machine learning systems that end up in front of people, not just in notebooks. I finished my B.Tech in Computer Science at Galgotias University in July 2026 and I am currently in the AI & ML programme at E&ICT Academy, IIT Roorkee. Most of what I know came from shipping: a decision support system that farmers could actually read, a job-search platform assembled from three services I had already built, and a knowledge base that answers questions about itself. I care about the part after the model works — the API, the interface, the deployment.
02 — What I Do
03 — Journey
Where I've been
Apr 2026 – NOW
AI & Machine Learning Programme
E&ICT Academy, IIT Roorkee · India
Jul 2026 – NOW
AI Second Brain
Personal project · Remote
Jul 2026 – Aug 2026
JobPilot
Personal project · Remote
Mar 2026 – Jun 2026
Full-Stack Developer Intern
Brokoders · Haldwani, Uttarakhand
Nov 2025 – Mar 2026
Tri-Model Agricultural DSS
Galgotias University · Greater Noida
Sep 2022 – Jul 2026
B.Tech, Computer Science & Engineering
Galgotias University · Greater Noida, Uttar Pradesh
04 — Work
Things I've built
01AI Second BrainAI / RAGin progressJul 2026 – PresentA knowledge management system that reads unstructured notes, files them against the PARA framework on its own, and answers questions about the whole collection in plain language.
- Ingests unstructured data and uses large language models to categorise content against the PARA framework, removing manual tagging.
- Retrieval-Augmented Generation pipeline with sentence-transformers vector embeddings, enabling semantic auto-linking and natural-language question answering over a personal knowledge base.
- Python
- LLMs
- RAG
- sentence-transformers
- Streamlit
02JobPilotPlatform / AutomationJul 2026 – Aug 2026A unified job-search platform built by merging three services I had already written into one system, with guardrails against the two ways this kind of tool usually goes wrong: hallucinated résumés and automated bulk email.
- Merged three independent microservices (harvester, résumé builder, email sender) into a monorepo using git subtree, preserving full commit history and consolidating them onto a single 14-table schema.
- Ethical web-scraping pipeline in Python with Playwright, aggregating and deduplicating postings from Naukri, RemoteOK, and Wellfound while enforcing rate limits and robots.txt compliance.
- AI résumé tailoring engine with deterministic guardrails against hallucination, guaranteeing truthful gap analysis and verified skill matching.
- Human-in-the-loop outreach using single-use, body-bound approval tokens to block automated bulk sending, with strict audit logging.
- Python
- Playwright
- PostgreSQL
- LLMs
- TypeScript
03Tri-Model Agricultural DSSML / ForecastingNov 2025 – Mar 2026A decision support system that pulls agronomic, climatic, and market data into one dashboard, runs three different models against it, and shows farmers why it said what it said.R² 0.90Crop yield — Random ForestF1 0.85Pest risk — Decision Tree2% MAPEPrice forecast — Prophet
- Synthesised agronomic, climatic, and market data into farming insights via a Streamlit dashboard.
- Developed and tuned three models: Random Forest regression for crop yield, Decision Tree classification for pest risk, and a Prophet time-series model for market price forecasting.
- Integrated a SHAP explainability engine so end users could interpret model outputs and act on them.
The problem
Farming decisions depend on three things that are normally forecast separately: how much a crop will yield, whether pests will hit it, and what the market will pay. Three tools with three interfaces is not a decision support system — it is homework. The goal was one dashboard where those three answers arrive together, in time to act on.
Why three different model families
Because they are three different problem shapes, and forcing one model onto all of them would have cost accuracy on at least two. Yield is continuous regression over tabular agronomic and climatic features, where Random Forest handles non-linear interactions without heavy feature engineering. Pest risk is classification, and a Decision Tree was chosen partly because its splits can be read back to a person — you can show the rule that fired. Price is a temporal series with seasonality, which is the problem Prophet exists for.
SHAP, and why it mattered here
A number a farmer cannot interrogate is a number a farmer will not act on. SHAP attributes each prediction back to the features that drove it, so the dashboard shows not just that yield will be X, but which conditions moved it there. That is the difference between a model output and a decision aid, and it is the part that took the longest to get right.
What the numbers mean
R² of 0.90 means the yield model explains 90% of the variance in held-out data. F1 of 0.85 balances precision against recall for pest risk, which matters because both error directions cost real money — a missed infestation and an unnecessary pesticide application are not symmetric mistakes, but neither is free. 2% MAPE means the price forecast lands within 2% of the actual price on average, which is the margin a planting decision can absorb.
- Python
- Scikit-Learn
- Prophet
- SHAP
- Streamlit
Knowledge graph
What the assistant knows
Every node is one chunk of the corpus the chat retrieves from, and every edge is a cosine similarity above 0.35 between two chunk embeddings. Nothing about the shape is drawn by hand — it is what the sentence-transformers/all-MiniLM-L6-v2 embeddings say, laid out by a force simulation over 35 nodes and 65 edges.
Drag a node to disturb the layout, or pick one to read what it holds.
- profile
- agricultural-dss
- jobpilot
- ai-second-brain
- faq
Read the graph as text (35 nodes)
- Gaurav Kashyap — profile · profile
Linked to: Is he available for work?, Journey, About, What he does, Certifications
- About · profile
Linked to: Gaurav Kashyap — profile, Journey, What he does, Certifications
- What he does · profile
Linked to: About, Skills, Gaurav Kashyap — profile, What works today
- AI Second Brain (AI / RAG, Jul 2026 – Present) · profile
Linked to: What works today, Relationship to this website, Status — IN PROGRESS, Skills, What is NOT built
- JobPilot (Platform / Automation, Jul 2026 – Aug 2026) · profile
Linked to: What it is, Two guardrails that shaped the design, The monorepo merge, Skills, Scraping, done deliberately
- Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026) · profile
Linked to: Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), What it does, The problem, Why three different model families, SHAP, and why it matters here
- Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026) · profile
Linked to: Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), Why three different model families, SHAP, and why it matters here, Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), The problem, What it does, What these metrics mean
- Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026) · profile
Linked to: What these metrics mean, What it does, Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026)
- Journey · profile
Linked to: Gaurav Kashyap — profile, About, Certifications, Education, Is he a recent graduate?, Tri-Model Agricultural Decision Support System, AI Second Brain, Is he available for work?, What professional experience does he have?
- Education · profile
Linked to: Journey, Is he a recent graduate?, Certifications, What professional experience does he have?
- Certifications · profile
Linked to: About, Journey, Education, Gaurav Kashyap — profile
- Skills · profile
Linked to: What he does, JobPilot (Platform / Automation, Jul 2026 – Aug 2026), AI Second Brain (AI / RAG, Jul 2026 – Present), What works today
- Tri-Model Agricultural Decision Support System · agricultural-dss
Linked to: Journey, JobPilot, AI Second Brain
- The problem · agricultural-dss
Linked to: Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), Why three different model families
- What it does · agricultural-dss
Linked to: Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), Why three different model families, Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), What these metrics mean
- Why three different model families · agricultural-dss
Linked to: Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), The problem, What it does, Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), SHAP, and why it matters here
- SHAP, and why it matters here · agricultural-dss
Linked to: Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), Why three different model families
- What these metrics mean · agricultural-dss
Linked to: Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), Tri-Model Agricultural DSS (ML / Forecasting, Nov 2025 – Mar 2026), What it does
- Repository · agricultural-dss
Linked to: Status — IN PROGRESS, Which projects have public code?, JobPilot
- JobPilot · jobpilot
Linked to: Tri-Model Agricultural Decision Support System, Repository, Which projects have public code?, What professional experience does he have?, AI Second Brain, Status — IN PROGRESS
- What it is · jobpilot
Linked to: JobPilot (Platform / Automation, Jul 2026 – Aug 2026), The monorepo merge, Two guardrails that shaped the design
- The monorepo merge · jobpilot
Linked to: JobPilot (Platform / Automation, Jul 2026 – Aug 2026), What it is, Two guardrails that shaped the design
- Scraping, done deliberately · jobpilot
Linked to: JobPilot (Platform / Automation, Jul 2026 – Aug 2026)
- Two guardrails that shaped the design · jobpilot
Linked to: JobPilot (Platform / Automation, Jul 2026 – Aug 2026), What it is, The monorepo merge
- AI Second Brain · ai-second-brain
Linked to: Tri-Model Agricultural Decision Support System, JobPilot, Status — IN PROGRESS, Journey
- Status — IN PROGRESS · ai-second-brain
Linked to: AI Second Brain (AI / RAG, Jul 2026 – Present), Repository, AI Second Brain, JobPilot, Which projects have public code?
- What works today · ai-second-brain
Linked to: AI Second Brain (AI / RAG, Jul 2026 – Present), Skills, What he does, Relationship to this website
- What is NOT built · ai-second-brain
Linked to: AI Second Brain (AI / RAG, Jul 2026 – Present)
- Relationship to this website · ai-second-brain
Linked to: AI Second Brain (AI / RAG, Jul 2026 – Present), How was this website built?, What works today
- Is he available for work? · faq
Linked to: Gaurav Kashyap — profile, Is he a recent graduate?, Journey
- Compensation · faq
No link above the 0.35 similarity floor.
- Is he a recent graduate? · faq
Linked to: Journey, Education, Is he available for work?
- What professional experience does he have? · faq
Linked to: JobPilot, Journey, Education
- Which projects have public code? · faq
Linked to: Repository, JobPilot, Status — IN PROGRESS
- How was this website built? · faq
Linked to: Relationship to this website
05 — Stack
What I work with
- Languages
- PythonJavaJavaScriptSQL
06 — Credentials
Education & certifications
EDUCATION
B.Tech, Computer Science & Engineering
Galgotias University
Sep 2022 – Jul 2026 · Greater Noida, Uttar Pradesh
CBSE Class XII — 80%
Delhi Public School
Mar 2022 · Ranchi, Jharkhand
CERTIFICATIONS
Artificial Intelligence and Machine Learning
E&ICT Academy, IIT Roorkee
Apr 2026 – Present
Claude Code 101
Anthropic
Aug 2026
Data Analytics Job Simulation
Deloitte Australia
Jun 2026
Prompt Design in Agent Platform
Google
Jun 2025
07 — Contact
Get in touch
Open to ML and full-stack roles · India · remote-friendly
- Emailgauravgk2244@gmail.com
- GitHubgithub.com/Gaurav-1008
- LinkedInlinkedin.com/in/gaurav-kashyapp
- LocationDelhi, India