Available for hireBengaluru, IndiaFounder CoreForge, GOLDEN's Coreforge. Unofficial dMAT practice tool at goldensdmat.in (opens in new tab)
- Gen AI & Data Science Engineer
- LLM Agent Architect
- Production ML Engineer
- Data Pipeline Builder
LLM fine-tuning · RAG · Multi-agent systems. Shipping production ML & agentic AI from Bengaluru.
01 / About
The story behind the stack.
CurrentlyAvailable for hire
Data Science & Gen AI Developer
- Multi-Agent Systems
- LLM Fine-tuning
- RAG
- Based in
- Bengaluru
- Open to
- Full-time & contract · Remote / Bengaluru
- Focus
- Production LLM agents & RAG systems
- Current program
- Agentic Systems and Design Program · iHUB DivyaSampark @ IIT Roorkee
- Education
- Master's in Data Science (Online MDS) · Expected Feb 2027
- Credentials
- 37 verified credentials
Bengaluru, India
Oikantik Basu
Data Science and Gen AI developer with hands-on experience in LLM fine-tuning, RAG systems, agentic AI architectures, and production ML deployments. Proficient in building and deploying solutions using LangChain, LlamaIndex, AutoGen, and leading LLM APIs (GPT, Claude, Bedrock, Gemini). Experienced in exploratory data analysis, data visualization, vector database optimization (Chroma/FAISS), and translating model outputs into actionable business insights. Skilled in Python, SQL, and cloud platforms (AWS, GCP), with a strong foundation in statistics and machine learning. Pursuing a Master's in Data Science from the University of Pittsburgh.
- Phone
- +91 7001124396
17
Public projects
50
Skills listed
5
Skill domains
70
Technologies in projects
On Kaggle
5-Day AI Agents Intensive Course with Google
Research Competitor
Code Submitter
Competitor
02 / Skills
Tools in the arsenal.
Spin it, or tap a node to lock it center-stage. Every one is something I've shipped production code with.Tap a node to open it. Every one is something I've shipped production code with.
Agentic AI9 skills
Data Science & ML12 skills
Analytics & Viz6 skills
Infrastructure9 skills
Try it — inline sentiment
Tiny lexicon model · runs in your browser · no API
Polarity
Label
positive
score 0.52
- fast: positive word, intensified ×1.6, weight 1.60
- accurate: positive word, weight 1.00
¬ negated · ×1.6 intensified
Optional: sends this text to Google Gemini for a second opinion.
Daily setup
Editor
VS Code · Cursor
OS
macOS · WSL
Shell
zsh · oh-my-zsh
Dev flow
GitHub · conventional commits
Containers
Docker · Compose
Keyboard
Keychron K2 · mono
03 / Projects
Things I've built.
From mechanistic-interpretability clinical AI to serverless voice commerce — 17 shipped projects.
17 projects
Closest toAI-generated · may be wrong
Founder · live productGOLDEN's Coreforge
CoreForge
dMAT Practice
Practice for the complete dMAT, the digitaler Mastertest for admission to German Master’s programmes: unlimited, freshly generated Core Module tasks, the General Academic Module and every subject module, all with real exam timing.
- TypeScript
- Supabase
Free plan — no card, no expiry
Unofficial practice tool. Not affiliated with g.a.s.t., TestDaF-Institut, APS, or the DAAD.
AI/ML · TypeScriptFeaturedAI-Native Clinical Ecosystem for Mechanistic Interpretability
A 30-agent, 8-layer clinical pipeline built on MedGemma + TxGemma + Gemini. Provides a Semantics-to-Vector interface so physicians can inspect how medical AI models reach their diagnoses — from lab report extraction and X-ray fracture detection to differential diagnosis and voice copilot.
- React 19
- Node.js
- PostgreSQL
- MedGemma
- TxGemma
- Gemini
- +2 more
Data Science · Jupyter Notebook20-Year Stock Analysis with ML, Deep Learning & Dashboard
Comprehensive data science project analyzing and forecasting TCS stock. Feature engineering with MACD/RSI/rolling stats, five models compared (Linear, Random Forest, XGBoost, LSTM, Prophet), and an interactive Streamlit dashboard with buy/sell signals via EMA crossover.
- Python
- Streamlit
- Plotly
- Scikit-learn
- +3 more
Data Science · Jupyter NotebookEnd-to-End Healthcare Data Science Platform
Complete ML pipeline analyzing drug safety, efficacy, and patient experiences from Drugs.com. Trained 6 ML models (classification, regression, clustering), performed association rule mining on drug-condition-side effect patterns, and built 3 interactive Streamlit dashboards.
- Python
- Streamlit
- Scikit-learn
- NLTK
- +3 more
AI/ML · PythonFeaturedServerless Agent Swarm for Stateful Orchestration
Async multi-agent orchestration engine built with Google ADK + Gemini 2.0 Flash + Cloud Run. Submit a complex task (e.g. analyze a 50-page PDF) and a coordinated swarm of specialized agents (Planner, Researcher, Writer, Reviewer) divides, conquers, and delivers — with real-time SSE streaming and per-agent cost tracking.
- Python
- FastAPI
- Google ADK
- Gemini 2.0 Flash
- Cloud Run
- +2 more
AI/ML · TypeScriptFeaturedAI Marketplace with WhatsApp Commerce & Voice-First Shopping
Omnichannel commerce for Indian retailers. Customers shop via WhatsApp or web chat in 8 Indian languages using voice notes. Gemini 2.0 transcribes voice, extracts product intents, routes to seller. Features Khata book OCR, Razorpay Route splits, and Bedrock-powered dead-stock detection.
- AWS Lambda
- DynamoDB
- Twilio
- Razorpay
- Bedrock
- +3 more
AI/ML · TypeScriptFeaturedEmpirical AI Tutor with Three Specialized Agents
Autonomous educational workspace powered by three AI agents: Teacher (Gemini 2.5 Flash Native Audio), Scientist (Gemini 3 Pro), Visualizer (Gemini 3 Flash). Upload any PDF → auto-generates slides, narrates via live voice, and lets you ask questions with real-time interruption. Generates videos via Veo.
- React 19
- TypeScript
- Vite 6
- Gemini 2.5 Flash Audio
- +3 more
AI/ML · TypeScriptFeaturedMultimodal AI Platform for Marketing Assets
Transforms rough inputs (text, images, voice notes) into polished marketing assets — copy, visuals, storyboards, voiceover scripts, and short promo videos. Includes Trend Analyzer for AI-powered trend discovery across Instagram Reels, X/Twitter, and LinkedIn. Built on GCP with Vertex AI Gemini.
- React
- Vertex AI
- Gemini 2.5 Flash/Pro
- Cloud Run
- +3 more
Data Science · Jupyter NotebookEmployee Attrition Analysis with Flask Dashboard
Predicts employee attrition using IBM's HR dataset (1470 × 35). Trained seven models including XGBoost (86% accuracy, 0.83 ROC-AUC), deployed via Flask web app that converts uploaded CSVs into dynamic Excel dashboards with automated HR recommendations.
- Python
- Scikit-learn
- XGBoost
- LightGBM
- +3 more
Data Science · Jupyter NotebookEnd-to-End Pipeline for 2008–2021 Netflix Catalog
Modular pipeline for Netflix's 14-year content catalog: PostgreSQL ingestion, Python cleaning and feature engineering, advanced EDA (time-series, heatmaps, wordclouds, statistical tests), association-rule mining, ML models, and an interactive Streamlit dashboard.
- Python
- PostgreSQL
- Pandas
- Streamlit
- +2 more
Web App · JavaScriptAI-Powered Quiz App with GitHub Cloud Sync
Modern MCQ quiz app built with vanilla JavaScript. Upload MCQ PDFs, get timed questions, and receive Google Gemini-powered performance analysis. Syncs scores across devices via a private GitHub Gist. Perfect for government job or certification prep.
- Vanilla JS
- HTML5
- CSS3
- Google Gemini API
- +2 more
AI/ML · TypeScriptMulti-Agent Nervous System for Large Sporting Venues
A Gemini 2.5 Pro orchestrator invokes five Gemini 2.5 Flash specialist agents (Flow, Queue, Concierge, Care and Revenue) through Google ADK’s AgentTool pattern to handle crowd movement, waiting times and real-time coordination. Every intervention lands in Firestore and recolours a React Three Fiber 3D twin of the venue, while a voice-enabled fan PWA answers each fan. Built solo in 48 hours for Hack2skill’s Virtual: PromptWars.
- Google ADK
- Gemini 2.5 Pro
- Gemini 2.5 Flash
- Cloud Run
- +5 more
AI/ML · PythonMultilingual Voter-Journey Agent for Indian Elections
A multimodal, multilingual voter-journey agent that collapses the Election Commission voter portal’s seven fragmented services into one conversation. Five Google ADK agents run on Cloud Run: Gemini 2.5 Flash routes each request, specialists handle registration, EPIC verification and polling-booth lookup, and a Gemini 2.5 Pro story agent explains why the voter’s own vote matters. Built solo for Hack2skill PromptWars 3 (Election Process Education).
- Google ADK
- Gemini 2.5 Flash
- Gemini 2.5 Pro
- FastAPI
- +6 more
AI/ML · PythonLangGraph Support Agent for a Lending-Operations Desk
The final capstone of Masai’s Banking & FinTech (Cred) track: a LangGraph support agent that answers loan-policy questions only from a 14-document knowledge base, with verbatim cited sentences and an “I don’t know” fallback, checks a loan application’s status through a tool with a designed escalation score, and refuses attempts to override its rules.
- Python
- LangGraph
- Chroma
- FastAPI
- +3 more
Data Science · JavaScriptCrime Analytics & Predictive Policing Assistant on Zoho Catalyst
Built for KSP Datathon 2026 on the official Karnataka State Police FIR database schema: an interactive geospatial command center that detects crime hotspots, uncovers criminal networks, links repeat offenders across jurisdictions, forecasts emerging risks and answers plain-English questions. Runs end to end on Zoho Catalyst over 45,000 synthetic FIRs.
- Zoho Catalyst
- React 18
- Vite
- Tailwind CSS
- +3 more
Data Science · JavaScriptMarket-Data Operations Dashboard with a Signal Wind Tunnel
A market operations dashboard on public Coinbase Exchange data for BTC, ETH and SOL, with an optional Confluent Cloud Kafka path. It helps an operations team spot stale feeds and unusually sharp observed price moves. Its Signal Wind Tunnel freezes a real observed sample, injects duplicates, missing or late events or a synthetic price shock, and compares unguarded and quality-aware decisions.
- Node.js
- JavaScript
- Kafka
- Confluent Cloud
- +1 more
Web App · TypeScriptAI Study Companion for Indian Government Exams
An AI-native study companion for Indian government and PSU exams. Pick a target exam and date and it builds an adaptive day-by-day plan from that exam’s real syllabus and pattern, starting with Coal India’s Management Trainee (Systems / CS) exam. Adding an exam means adding taxonomy data, not code.
- React
- TypeScript
- Vite
- Tailwind CSS
- +4 more
Web App · TypeScriptCinematic Showcase Site for a Kolkata Street-Food Shop
A cinematic, mobile-first showcase site for The Calcutta Classics, a street-food shop in Shyambazar, Kolkata: a video hero, a React Three Fiber momo, a filterable menu with fuzzy search, playful foodie tools, and content in English and Bengali. A showcase, not an ordering site: no cart and no checkout.
- Next.js 16
- TypeScript
- Tailwind CSS v4
- GSAP
- +2 more
04 / Founder · GOLDEN's Coreforge
I built CoreForge.
Practice for the complete dMAT, the digitaler Mastertest for admission to German Master’s programmes: unlimited, freshly generated Core Module tasks, the General Academic Module and every subject module, all with real exam timing.
Founder of GOLDEN's Coreforge · since Jul 2026goldensdmat.in (opens in new tab)
Try 10 free questions (opens in new tab)Start free (opens in new tab)Do I need the dMAT? (opens in new tab)
Free plan — no card, no expiry. See pricing (opens in new tab)
Homegoldensdmat.in/welcome (opens in new tab) 
The daily demo on a phonegoldensdmat.in/demo (opens in new tab) 
Brain Gymgoldensdmat.in/brain-gym (opens in new tab) 
Every subject modulegoldensdmat.in/modules (opens in new tab)
- subject modules, every one in the official table
- 8
- Core Module subtests, each at exam timing
- 3
- per Core subtest of 20 tasks
- 25:00
- the real subject-module slot
- 90 min
- free demo questions a day, no account needed
- 10
- Brain Gym drills, ten levels each
- 40
What CoreForge does
Free daily demo (opens in new tab)
10 originally generated questions a day at real exam timing. No account, no card.
Generated, not stored (opens in new tab)
Unlimited, freshly generated Core tasks in the official formats, each checked by its generator before it is served.
Full timed Core Module, on the Pro plan (opens in new tab)
Three Core subtests of 20 tasks in 25:00 each, building up to the full sitting of about 3.5 hours.
Every subject module, on the Pro plan (opens in new tab)
The General Academic Module across all 8 topic areas, Computer Science, five Battery Science tracks and Data Science (modeled format).
Brain Gym (opens in new tab)
Forty short, seeded drills for speed, attention, memory and reasoning, ten levels each. One round a day needs no account.
Do I need the dMAT? (opens in new tab)
Type your degree and see it matched against APS India’s official list of affected fields.
Which module do I need? (opens in new tab)
Three questions about your target programme and APS route, checked against the official module table.
Feedback and review (opens in new tab)
Instant feedback after each answer and a full answer review; Pro adds Learn, analytics and a mistakes notebook.
Installable, offline-ready (opens in new tab)
Installs as an app with offline support, and a Supabase backend syncs your practice across devices.
Seven dMAT guides (opens in new tab)
Seven guides on registration through g.a.s.t., timing, scoring, which module you sit and the APS India process.
Follow the marker
Four frames, one rule. Pick the cell the marker lands on next.
Where does the marker land in frame 5?
Loading today’s pattern…
A small puzzle I wrote for this page in the style of a figure sequence, not a CoreForge question: CoreForge’s own tasks are harder and timed. A new pattern every day, generated in your browser, and only shipped when every rule that fits the four frames agrees on the fifth.
How the questions are made
The 3 Core subtests are generated, not stored: every task is freshly generated in the official format and checked by its generator before it is served.
Figure Sequences. Passes a strict inferability check: the rule must be uniquely deducible.
Mathematical Equations. Brute-forced to guarantee exactly one whole-number solution.
Latin Squares. Verified so every valid completion agrees on the answer cell.
General Academic Module questions keep the official shape, four options with exactly one correct, and each one passes a validator before it ships.
Built with
- TypeScript · Question-generation algorithms, built with Claude
- Supabase · Backend with cross-device sync and leaderboards
Free to start
The Free plan never expires and needs no card; Pro unlocks the rest.
Free plan
Free plan — no card, no expiry
- Core Module practice: Figure Sequences, Mathematical Equations, Latin Squares (easy, 3 per set)
- Your results, answer review and history
- Brain Gym levels 1–3
- No card, no expiry
CoreForge Pro
Current prices are on the pricing page.
- Every difficulty and set size, plus the full timed Core Module
- General Academic Module passages across all 8 topic areas
- Every subject module in the official table
- Unlimited mock exams and the full dMAT simulation
- Learn, analytics, mistakes notebook, weekly leagues
- AI tutor
- Monthly or yearly, cancel in one tap
- 7-day money-back guarantee on your first purchase
Unofficial practice tool. Not affiliated with g.a.s.t., TestDaF-Institut, APS, or the DAAD. Question formats follow publicly documented dMAT task types; all questions are originally generated. dMAT is a trademark of its owners.
05 / Experience
The journey so far.
Internships, freelance AI work, and a Master's in progress — all pulling in the same direction.
Owner
GOLDEN's Coreforge
goldensdmat.intoPresent4 monthsIndia · On-site- Founded GOLDEN's Coreforge and built goldensdmat.in for dMAT practice
- Question generators in TypeScript with Claude: unique, verifiable tasks
- Supabase backend with cross-device sync and leaderboards
CoreForge · dMAT Practice
Practise the complete dMAT with real exam timing.
Unofficial dMAT practice tool
Founded GOLDEN's Coreforge and developed goldensdmat.in, a practice platform for the dMAT (digitaler Mastertest) for students applying to German Master's programmes. Engineered question-generation algorithms in TypeScript using Claude so every task is unique and verifiable, authored a content bank of 295 questions across academic fields, and built the backend on Supabase with cross-device sync and leaderboards.
- 295 Questions authored
- ProgramPresent
Agentic Systems and Design Program
iHUB DivyaSampark @ IIT Roorkee
toPresent10 monthsRemote- Multi-agent systems on ReAct + LangGraph
- RAG pipelines with Chroma/FAISS optimization
- LLMOps: tracing, evaluation, FastAPI serving
Skills used (each opens it in the Skills section)
Architecting multi-agent systems using ReAct framework and LangGraph for autonomous reasoning. Building advanced RAG pipelines with LlamaIndex, LangChain, and vector databases (Chroma/FAISS), optimizing retrieval accuracy through semantic chunking and vectorization. Implementing LLMOps best practices for production deployment including agent observability (tracing), evaluation frameworks, and REST API serving via FastAPI.
- 4+ Agent frameworks
- 2× Vector DBs tuned
- FastAPI Serving layer
- FreelancePresent
Freelance AI Agent Specialist
Mindrift
toPresent1 year 5 monthsRemote- RLHF evaluations for GPT + Claude alignment
- AutoGen multi-agent code evaluation
- Hallucination detection & eval metrics
Skills used (each opens it in the Skills section)
Conducting RLHF evaluations to align LLMs (GPT, Claude) with safety and accuracy guidelines. Evaluating complex Python code generation and reasoning traces using AutoGen-based multi-agent workflows. Identifying and documenting hallucinations and logic failures in state-of-the-art models, building evaluation metrics that improved model reliability.
- GPT + Claude LLMs aligned
- AutoGen Eval workflow
- RLHF Feedback loop
- Freelance
Freelance · Project Scatter
Alignerr
to3 monthsRemote- Reusable 2D and 3D scatter functions with cluster annotations
- Seaborn and Plotly output with label toggles and hover tooltips
- Vectorized plotting for large datasets, shared as a team-wide module
Skills used (each opens it in the Skills section)
Developed an advanced version of Project Scatter, a custom visualization toolkit in Python for interactive scatter plots of multidimensional datasets: reusable 2D and 3D scatter functions with cluster annotations, static and interactive output with Seaborn and Plotly, real-time correlation insights, label toggles and hover tooltips, and vectorized plotting for large datasets. Shared as a Python module for team-wide use.
- Freelance
Freelance AI Trainer
Outlier
to2 monthsRemote- Designed and evaluated model training data
- Refined model outputs for ethical AI behavior
- Evaluation rubrics for accuracy and reliability
Designed and evaluated training data to enhance machine learning model performance. Collaborated with cross-functional teams to refine model outputs and ensure ethical AI behavior, and developed evaluation rubrics to improve the accuracy and reliability of AI in real-world applications.
- Internship
Data Science Intern
Unified Mentor Private Limited
to4 monthsGurgaon, Haryana- EDA + regression on IBM HR Analytics
- SQL data mining for TCS stock time-series
- Interactive visualization dashboards
Skills used (each opens it in the Skills section)
Performed exploratory data analysis and regression modeling on IBM HR Analytics dataset, uncovering key attrition drivers via interactive visualizations. Built SQL-based data mining queries to extract and transform structured datasets, delivering actionable insights on TCS stock trends through time-series analysis.
- 86% HR attrition accuracy
- 20y TCS stock analyzed
- R² 0.999 Trend fit
WordPress Developer
Telangana Desam Leader
to1 year 3 monthsRemote- Built an awareness-focused blog website
- Managed weekly updates
Developed a blog website focused on spreading awareness and managed its weekly updates.
- Freelance
Freelance WordPress Developer
K.pop Merchandise
to1 year 2 months- WordPress and Elementor Pro sites with custom CSS
- Power BI and Excel analysis for Shiprocket operations
- B2B business processes
Skills used (each opens it in the Skills section)
Built websites using WordPress and Elementor Pro with custom CSS. Used Power BI and Excel for data entry and analysis for Shiprocket operations, and worked on B2B business processes.
Hackathons & launches
- Founder
Founded GOLDEN's Coreforge — goldensdmat.in
A dMAT (digitaler Mastertest) practice platform for students applying to German Master's programmes, free to start: generators for figure sequences (with an inferability check), equation systems (brute-forced to a unique whole-number solution) and Latin squares (verified unique answer cell), real exam timing, a Supabase backend with cross-device sync and leaderboards, and a 295-question content bank at launch.
- Build
Sankalp — PromptWars 3, Election Process Education track
Run by Hack2skill
Multimodal voter-journey agent: EPIC card OCR with Cloud Vision, multilingual voice, five Google ADK agents (an orchestrator routing to four specialists), Gemini 2.5 Flash routing and Gemini 2.5 Pro narratives, and Firestore sessions on Cloud Run (asia-south1).
- Finalist
Top 36 Finalist — AI for Bharat Hackathon
Run by Amazon Web Services (AWS) · Hack2skillTeam of two
VyaparGyan, an agentic orchestration platform for Bharat's retailers built with generative AI on AWS, was selected as a Top 36 Finalist for the grand finale (April 6-7).
- Submission
ShieldFlow — GitLab Hackathon 2026 entry
Run by GitLab
Autonomous security and compliance review pipeline for GitLab merge requests: webhook-triggered agents for change risk, threat modelling, compliance mapping (SOC 2, GDPR, HIPAA), remediation, energy hotspots and a Claude code review, posting a structured comment back to the merge request. Python 3.12, FastAPI, Cloud Run, Gemini and Claude.
- Build
PULSE Stadium AI — Virtual: PromptWars
Run by Hack2skill
Multi-agent operating layer for a 40,000-seat stadium: a Gemini 2.5 Pro orchestrator composing five Gemini 2.5 Flash specialist agents through Google ADK, a React Three Fiber digital twin, a voice-enabled fan PWA, five live Cloud Run services and a counterfactual simulator. Built solo in 48 hours.
- Submission
Submission — The MedGemma Impact Challenge (Kaggle)
Run by Kaggle · Google Research
UrbanCare AI was submitted to Kaggle's MedGemma Impact Challenge, hosted by Google Research: a 30-agent, 8-layer clinical pipeline built on MedGemma 4B-it and TxGemma 9B via Hugging Face Inference Endpoints, plus Gemini.
Education
Master's in Data Science (Online MDS)
University of Pittsburgh (Coursera)
to
Expected Feb 2027
B.Tech in Computer Science & Engineering
Techno International New Town, Kolkata
to
CGPA: 8.22
Loading…
Loading…
What I'm reading
Papers and books that shape how I build
6 of 6 shown
- Foundations · 2017arXiv:1706.03762
Attention Is All You Need (opens in new tab)
Vaswani et al.
Transformer architecture — the starting point for every LLM I work with.
- RAG · 2020arXiv:2005.11401
Retrieval-Augmented Generation for Knowledge-Intensive NLP (opens in new tab)
Lewis et al.
The original RAG paper. Re-reading it while tuning my Chroma/FAISS pipelines.
- Agents · 2022arXiv:2210.03629
ReAct: Synergizing Reasoning and Acting in Language Models (opens in new tab)
Yao et al.
Backbone of the multi-agent work I'm shipping at iHUB.
- RAG · 2023arXiv:2307.03172
Lost in the Middle: How Language Models Use Long Contexts (opens in new tab)
Liu et al.
Why chunk order and re-ranking matter more than raw context length.
- Alignment · 2022arXiv:2212.08073
Constitutional AI: Harmlessness from AI Feedback (opens in new tab)
Anthropic
Direct influence on the RLHF eval workflows I run at Mindrift.
- Systems · 2017
Designing Data-Intensive Applications (opens in new tab)
Martin Kleppmann
Ground truth for building reliable backends under any LLM layer.
06 / Certifications
Credentials you can verify.
Every badge and certificate below links to the issuer's own verification page.
37verified credentials
From 5 issuers on 3 platforms. links checked
- 20 course-completion badges
- 16 course certificates
- 1 professional certificate
Professional certificateIssued by Google on Coursera
Google AI Professional Certificate
Issued , 7 courses
- Verify (opens in new tab)
AI Fundamentals
Course certificate, issued
- Verify (opens in new tab)
AI for Brainstorming and Planning
Course certificate, issued
- Verify (opens in new tab)
AI for Research and Insights
Course certificate, issued
- Verify (opens in new tab)
AI for Writing and Communicating
Course certificate, issued
- Verify (opens in new tab)
AI for Content Creation
Course certificate, issued
- Verify (opens in new tab)
AI for Data Analysis
Course certificate, issued
- Verify (opens in new tab)
AI for App Building
Course certificate, issued
Claude Academy
20 course-completion badges
Course-completion badge
AI Fluency for pK–12 educators
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
AI Fluency for nonprofits
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
AI Fluency for small businesses
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
AI Fluency for students
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Building effective human-agent teams (beta)
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Building with the Claude API
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
AI Fluency: Framework & Foundations
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
AI Fluency for educators
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
AI Capabilities and Limitations
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
AI Fluency for Creative Work
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
AI Fluency for Builders
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Claude with Amazon Bedrock
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Claude with Google Cloud's Vertex AI
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Claude Code in Action
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Deploying Claude Enterprise with Confidence: The five decisions that shape your rollout
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Claude Platform 101
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Introduction to Model Context Protocol
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Claude Code 101
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Introduction to Claude Cowork
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)Course-completion badge
Claude 101
Issued by Anthropic on Claude Academy
Issued
Verify (opens in new tab)
Coursera
7 course certificates
Course certificate
Databases and SQL for Data Science with Python
Issued by IBM on Coursera
Issued
Verify (opens in new tab)Course certificate
Understanding and Visualizing Data with Python
Issued by University of Michigan on Coursera
Issued
Verify (opens in new tab)Course certificate
Python Project for Data Science
Issued by IBM on Coursera
Issued
Verify (opens in new tab)Course certificate
Python for Data Science, AI & Development
Issued by IBM on Coursera
Issued
Verify (opens in new tab)Course certificate
Data Science Methodology
Issued by IBM on Coursera
Issued
Verify (opens in new tab)Course certificate
Tools for Data Science
Issued by IBM on Coursera
Issued
Verify (opens in new tab)Course certificate
What is Data Science?
Issued by IBM on Coursera
Issued
Verify (opens in new tab)
Programiz PRO
2 course certificates
Course certificate
Learn Python Basics
Issued by Programiz on Programiz PRO
Issued
Verify (opens in new tab)Course certificate
Practice: Python Basics
Issued by Programiz on Programiz PRO
Issued
Verify (opens in new tab)
07 / Kaggle
On Kaggle.
What I'm competing in right now, the writeups I've published and the badges I've earned, read from Kaggle and limited to what is public.
- OverallContributor
- CompetitionsContributor
- DatasetsContributor
- NotebooksContributor
- Badges earned
- 20
- Writeups published
- 2
- Competitions entered
- 14
Updated · live from Kaggle
Currently competing
- Featured
Closes 12 Nov 2026
Google - The Gemma 4 Developer Agent Paper Track
Hosted by Google DeepMind
Submit a research paper advancing agentic software engineering.
Deadline
244 teams on the leaderboard
- Featured
Closes 2 Dec 2026
Google - The Gemma 4 Developer Agent Competition
Hosted by Google DeepMind
Accelerate research in autonomous coding agents.
Deadline
Rank 60 of 3,043 teams · as of 17 Oct 2026
- Featured
Closes 14 Dec 2026
Enveda CASMI 2026 - Molecule ID From Mass Spectra
Hosted by Enveda
Predict 2D chemical structures of small molecules detected in complex biological extracts
Deadline
3,472 teams on the leaderboard
Writeups
Hackathon projectPublished
UrbanCare AI: A Multimodal Clinical Copilot for Real-Time Hospital Intelligence
A real-time hospital AI system combining vision, voice, and medical reasoning for collaborative patient care
Written for The MedGemma Impact Challenge
Modern healthcare systems impose an enormous administrative and cognitive burden on physicians. Doctors spend more time fighting clunky EMR interfaces, manually parsing unstructured lab reports, and writing repetitive…
Hackathon projectPublished
Unlocking Transparent Intelligence: Fine-Tuning Gemma 3 for Verifiable Reasoning
A GRPO-based approach to Chain-of-Thought reasoning on TPU v5e
Written for Google Tunix Hack - Train a model to show its work
In critical domains like mathematics, science, and coding, an AI model's answer is only as valuable as the logic behind it. Standard Large Language Models (LLMs) often suffer from the "Black Box" problem—delivering…
Badges
5-Day AI Agents Intensive Course with Google
Earned
Completed the 5-Day AI Agents Intensive Course 2025. Participants attended daily seminars, studied white papers, completed daily assignments, and created a capstone project about AI agents.
Research Competitor
Earned
Made a submission to a Research competition. Research competitions feature problems which are more experimental, may not have a clean or easy solution, and which are integral to a specific domain.
Code Submitter
Earned
Made a submission to a code competition. These competitions require the submission of a reproducible bundle of code rather than a simple file of predictions.
Competitor
Earned
Made a submission to a Kaggle competition that is eligible for points or medals. These competitions are full-scale machine learning challenges which pose difficult real world prediction problems.
Code Tagger
Earned
Added tags to a notebook. Tagging a notebook makes it more discoverable across Kaggle.
Vampire
Earned
Turned on the Kaggle dark theme. You can make this change under settings. Dark mode is a popular alternative color scheme for the site which many people find easier on the eyes.
- Biohub - Cell Tracking During Development (opens in new tab)
Biohub · Research
Closed · 3,947 teams
Rank 137 of 3,947 teams · as of 17 Oct 2026
- NVIDIA Nemotron Model Reasoning Challenge (opens in new tab)
NVIDIA · Featured
Closed · 4,185 teams
- The Gemma 4 Good Hackathon (opens in new tab)
Google DeepMind · Featured
Closed · 1,604 teams
- Measuring Progress Toward AGI - Cognitive Abilities (opens in new tab)
Google DeepMind · Featured
Closed · 1,062 teams
- AI Mathematical Olympiad - Progress Prize 3 (opens in new tab)
AI|MO · Featured
Closed · 4,138 teams
- March Machine Learning Mania 2026 (opens in new tab)
Kaggle · Featured
Closed · 3,462 teams
Rank 1,049 of 3,462 teams · as of 17 Oct 2026
- The MedGemma Impact Challenge (opens in new tab)
Google Research · Featured
Closed · 869 teams
Writeup published - Google Tunix Hack - Train a model to show its work (opens in new tab)
Google Cloud · Featured
Closed · 319 teams
Writeup published - Google DeepMind - Vibe Code with Gemini 3 Pro in AI Studio (opens in new tab)
Google DeepMind · Featured
Closed · 4,083 teams
- Agents Intensive - Capstone Project (opens in new tab)
Kaggle · Community
Closed · 11,455 teams
- BigQuery AI - Building the Future of Data (opens in new tab)
Google Cloud · Featured
Closed · 276 teams
08 / Principles
How I build.
Six working beliefs I return to — earned from production LLM systems, not from reading about them.
- 01
Ship the boring version first.
A working end-to-end pipeline beats a half-finished clever one. Prove the wire, then tune the signal.
- 02
Evals before vibes.
If you can’t measure it, you’re guessing. Every LLM feature gets a scored test set before it gets a UI.
- 03
Small models, sharp prompts.
Most production problems don’t need a frontier model — they need a well-indexed retrieval layer and crisp instructions.
- 04
Observability is a feature.
Traces, token counts, latency histograms, cost dashboards. The team that can see the system can fix the system.
- 05
Data > architecture.
A clean, labeled, deduped dataset beats a fancy model. Spend the weekend on the CSVs, not the ConvNet.
- 06
Write the docs you wish existed.
Half of engineering is unblocking the next person — usually future you. README first, commit second.
09 / Contact
Let's build something.
Open to research, full-time AI/ML roles, and ambitious freelance work. I usually reply within a day.
