AI/ML Portfolio · Applied Systems

AI/ML
Projects
Ecosystem

End-to-end ML pipelines, deployed AI systems, and cross-platform products — from transformer theory to production inference.

Scroll
4
GitHub Repos
15+
Distinct Projects
94.96%
Scam Detector Acc.
517
Academy Topics

Technical Stack

Built from
first principles

Every project is engineered end-to-end — mobile apps, browser automation, multi-agent platforms, and theoretical ML research — with a preference for self-healing, robust systems that handle model output unpredictability gracefully.

Stack: PyTorch · HuggingFace Transformers · scikit-learn · NumPy for ML, Cloudflare Workers · FastAPI · Flask · Supabase for infrastructure, Kotlin · Flutter · Electron · Playwright for cross-platform delivery.

PyTorchHuggingFaceCloudflare WorkersKotlinFlutterElectronK2-Think-v2SupabaseFastAPI

Monorepos & Ecosystems

GitHub Profile →
Repo 01 — Flagship
AI Systems Ecosystem

Stremini AI + KhetiAI

Full monorepo: Stremini AI 15-agent workspace, AI keyboard (StreminiIME), Electron desktop automation agent, Stremini Academy (517 topics, 25 tracks, 6 tiers), and KhetiAI crop disease detection with APMC mandi price forecasting.

Cloudflare WorkersK2-Think-v2KotlinFlutterElectronPlaywright
Repo 02 — ML/DL
Applied AI & ML

Applied AI & ML Projects

DistilBERT IT-support fine-tuning (86.7% acc.), character-level LSTM in pure NumPy, modular CNN on CIFAR-10 (87.3% acc.), 4-layer neural net from scratch, and a Prompt Engineering Lab with automated LLM-as-judge scoring pipeline.

PyTorchHuggingFaceNumPyscikit-learnBPTT
Repo 03 — Research
Research Work

Research Work Ecosystem

ML Scam Detector pipeline (94.96% accuracy, 28k URLs), interactive HTML papers on Gradient Descent as a Dynamical System and Spectral Properties of Attention Matrices, plus a FastAPI inference endpoint for the deployed model.

scikit-learnFastAPISHAPMathJaxTF-IDF
Repo 04 — Products
AI Projects Ecosystem

STYFI · Scholira · Zz-AI

Three full-stack AI products: STYFI fashion AI backend (Gemini 2.5 Flash + two-stage SDXL enhancement), Scholira scholarship RAG engine (110 scholarships, TF-IDF from scratch), and Zz-AI desktop voice agent with tiered LLM dispatch.

React + ViteFlutterElectronGemini 2.5SDXLRAG
Repo 05 — SaaS
Restaurant SaaS

OrbitDine

Restaurant management SaaS with React + TypeScript + Vite + Supabase. RLS policy consolidation, multi-tenant auth, post-signin null restaurant state resolution, and production-ready role-based access control.

ReactTypeScriptSupabaseViteRLS
Portfolio — Live
Research Portfolio

Research Portfolio Site

Live portfolio housing interactive papers, case studies, project writeups, and a trust score demo deployed as a live Cloudflare Worker endpoint.

HTML / CSSVercelCloudflare Workers

Detailed Breakdowns

01
Flagship AI Ecosystem15 Agents · 4 Platforms · Deployed

Stremini AI — Cross-Platform Multi-Agent Workspace

Production AI ecosystem: 15-agent web workspace (Intelligence, Creation, Strategy, Execution layers), Android AI keyboard (StreminiIME), desktop Electron automation agent with observe-act loop, and Stremini Academy — 517 topics across 25 tracks and 6 tiers.

  • Token-streaming SSE inference pipeline with think-block gating, in-flight promise deduplication, and retry-with-backoff — ~60% reduction in perceived latency
  • Map-reduce document QA supporting 600k-character inputs with parallel chunk extraction (batches of 4), NOT_RELEVANT filtering, and a final synthesis pass
  • AI keyboard backend with 7 endpoints (autocomplete, tone rewrite, grammar, translation 40+ languages) — dual-model routing: K2-Think-v2 for reasoning, Groq Llama 4 Scout for translation
  • Robust JSON repair pipeline: stripReasoning(), repairJSON(), extractJSON() — handling K2-Think-v2 think-block output quirks in production
Cloudflare WorkersK2-Think-v2KotlinFlutterElectronPlaywrightGroq Llama 4 ScoutSSE Streaming
View Repo →
02
AgriTech · ImpactDual-Model · React 19

KhetiAI / KhetiPlus — Agricultural AI for Indian Farmers

AI-powered crop intelligence for underserved Indian farmers. Dual-model architecture — Gemini 2.5 Flash for vision, K2-Think-v2 for reasoning — on a React 19 + Cloudflare Worker stack.

  • Crop disease detection returning structured pathology reports (diagnosis, causal agent, treatment) from base64 image input via Gemini vision API
  • 3-month APMC mandi price forecasting in JSON mode with trend classification: rising / stable / falling
  • Multi-agent Intelligence Hub covering pest migration patterns and soil health analysis
React 19Gemini 2.5 FlashK2-Think-v2FlutterCloudflare WorkersSupabase
View Repo →
03
AI Security · ML Pipeline94.96% Accuracy · 28k URLs

ML Scam Detector — Interpretable Security Pipeline

End-to-end interpretable scam detection: 16 hand-crafted features with Logistic Regression + Platt calibration, and a malicious URL detector trained on 28,000 URLs (URLhaus + Tranco Top 1M). Deployed as live Cloudflare Worker.

  • 100% recall, 95% precision on 1,000-sample stratified dataset with SHAP-analogous attribution and calibration theory
  • Three-tier ensemble: heuristics to ML to LLM with confidence-based routing — cutting LLM inference cost by ~90%
  • Live trust score endpoint: websitetrustscore.vishwajeetadkine705.workers.dev
scikit-learnLogistic RegressionSHAPFastAPIPlatt CalibrationURLhaus
View Repo →
04
Multi-Product EcosystemFashion · EdTech · Desktop

STYFI · Scholira · Zz-AI

Three full-stack AI products. STYFI: fashion AI backend with Gemini 2.5 Flash + two-stage SDXL enhancement. Scholira: scholarship retrieval with TF-IDF from scratch (110 scholarships). Zz-AI: voice-first desktop agent with tiered LLM dispatch.

  • STYFI: outfit intelligence via Gemini 2.5 Flash, segmentation + SDXL img2img producing Amazon/Myntra-quality imagery
  • Scholira: TF-IDF built from scratch — tokeniser, IDF, cosine similarity, no external ML lib. Two-stage ranking with domain-specific boosts
  • Zz Voice OS: regex intent to LRU cache to daily token quota to Groq Llama 3.3-70b to fallback. HS256 JWT and Supabase REST client both from scratch
React + ViteFlutterElectronGemini 2.5 FlashSDXLTF-IDF (scratch)Groq Llama 3.3
View Repo →

Benchmarks & Implementations

View all in repo →
CNN Pipeline

CIFAR-10 CNN — Modular Pipeline

1.2M parameter CNN with batch normalisation, OneCycleLR scheduler, and label smoothing. Fully modularised: data, model, train, evaluate, visualize modules.

87.3% test acc · 30 epochs
From Scratch · Deep Learning

Character-Level LSTM (Pure NumPy)

Full LSTM in pure NumPy: all gates, BPTT, gradient clipping, and Adam. Temperature-scaled text generation. Weights serialised as .npz with loss history tracking.

Zero deep learning framework
From Scratch · Fundamentals

4-Layer Neural Net (Pure NumPy)

Forward pass, backpropagation, mini-batch SGD (batch=32, 500 epochs). All matrix ops explicit: activations, cross-entropy loss, live loss/accuracy curves.

Zero autograd
Prompt Engineering · LLM Research

Prompt Engineering Lab

8 controlled A/B experiments isolating few-shot, CoT, persona, format constraints, and task decomposition vs zero-shot baseline. Automated LLM-as-judge pipeline with 4-axis rubric on a 1-10 scale using K2-Think-v2. Exportable as JSON/Markdown.

8 experiments · K2-Think-v2

Full Stack

ML & Research
PyTorchHuggingFace Transformersscikit-learnNumPyBPTTSHAPTF-IDFPlatt Calibration
LLM & Inference
Prompt EngineeringSSE StreamingThink-block ExtractionMulti-Agent OrchestrationJSON RepairLRU CachingRAG Pipelines
Infrastructure
Cloudflare WorkersFastAPIFlaskSupabaseElectronKotlinFlutterPlaywright
MBZUAI
BSc Artificial Intelligence Applicant · 2027
Stanford University
Machine Learning Certification
DeepLearning.AI
Deep Learning Specialization
HuggingFace Hub
Published: DistilBERT IT Support Classifier

Get in Touch

Collaborate on
what matters

Open to ML research collaborations, AI safety discussions, internship opportunities, and conversations about building technology for underserved communities.

Context & Background

Models in Production

MBZUAI K2-Think-v2 · Gemini 2.5 Flash · Groq Llama 4 Scout · DistilBERT · Groq Llama 3.3-70b

Research Interests

Spectral theory of neural networks · AI safety and prompt injection defence · Interpretable ML for security · Multi-agent architectures · Agricultural AI for South Asia

Collaboration Areas

AI for social impact in South Asia · Educational AI platforms · AI safety tooling · Applied ML research with deployable outcomes

Languages

Python · JavaScript / TypeScript · Kotlin · Dart (Flutter)