Skip to content
PP Logo
Open to opportunitiesPratham Pansare

FrontEnd BackEnd Developer

I won't bore you with walls of text — the work below does the talking. I build intelligent systems, AI-powered applications, and full-stack products — and I ship them end-to-end: the model, the API, and the interface it deserves.

Selected WorkContact

/ Projects

01 / 07 — INDEX

Seven systems, built end-to-end — from research-grade ML pipelines to production interfaces. Each one turns a hard problem into working software.

AI SECURITY(2026)

TGDetect

Temporal Graph Threat Detection

A cybersecurity threat detection and investigation system built on temporal heterogeneous graphs — pairing a GraphSAGE + GRU pipeline with an interactive analyst workspace for hunting botnet behaviour in raw network telemetry.

ROC-AUC
0.9999
Events
130K
Latency
<200ms
  • ROC-AUC 0.9999 on held-out CTU-13 botnet data
  • ≈130K events processed in under 200ms via zero-copy Parquet streaming
  • Interactive temporal graph visualization
  • Force-directed physics layout
  • Threat-prioritized subgraph sampling
PythonPyTorchFastAPIGraphSAGE
Live DemoSource
LLM TOOLING(2026)

Verbis

Intelligent AI Database Assistant

An AI-powered database assistant that combines natural-language-to-SQL generation, schema exploration, and privacy-aware handling — shipped as a terminal REPL and a VS Code extension with multi-provider LLM support.

Providers
3 LLMs
Surface
CLI + IDE
Guard
Read-only
  • Terminal REPL for database conversation
  • Natural language to SQL generation
  • Privacy-preserving schema anonymization
  • Gemini, Groq, and Ollama integration
  • OS keychain security + read-only execution guards
  • Live ER diagram and query history visualization
  • Business glossary for domain context
TypeScriptPythonFastAPIVS Code ExtensionReact
Live DemoSource
COMPUTER VISION(2025)

ShetVaidya

AI Crop Disease Detection

An AI-powered crop disease inference and medicine verification platform engineered for real-world agricultural use — confidence-gated MobileNetV2 diagnosis, non-leaf rejection, and batch medicine verification in a mobile-first workflow.

Model
MobileNetV2
Platform
Mobile-first
Storage
Cloudflare R2
  • MobileNetV2 disease inference engine
  • Confidence gating + non-leaf image rejection
  • Medicine verification with barcode scanning
  • Batch verification workflow
  • Mobile-first field UI
  • Cloudflare R2 image storage
ReactFastAPITensorFlowMobileNetV2
Live DemoSource
RESEARCH TOOLING(2026)

Kivo

Research Discovery Workspace

A research workspace for scholarly discovery and literature review workflows — federated academic search, evidence matrices, and rigorous source tracking built for people who read papers seriously.

Sources
3 APIs
Tracking
SHA-256
Export
BibTeX / RIS
  • Federated search across Crossref, arXiv, and Europe PMC
  • Evidence matrix for structured synthesis
  • Research comparison workflows
  • PDF extraction with passage capture
  • SHA-256 source integrity tracking
  • BibTeX / RIS export
  • Publication monitoring
TypeScriptNext.jsPostgreSQLPrisma
Live DemoSource
ANDROID ENGINEERING(2025)

Tenora

Offline-First Property & Rent Accounting

An offline-first Android application for property management and rent accounting — rooms, tenants, month-by-month ledgers, partial payments, PDF receipts, reports, and spreadsheet imports with core accounting and storage running entirely on the device.

Storage
On-device
Backup
AES-GCM
Receipts
PDF
  • Month-by-month rent ledger with derived paid / partial / unpaid standing
  • Partial and multi-month payments with allocation preview
  • Transactional payment editing and reversal
  • PDF rent receipts with WhatsApp sharing
  • CSV/XLSX import with validation preview and rollback
  • Collection, outstanding, and defaulter reports
  • AES-GCM encrypted local backups
KotlinJetpack ComposeRoom / SQLiteMaterial 3
Source
PRIVACY TOOLING(2026)

Veyla

Private Android Workspace Browser

A native Android workspace browser focused on organized, private, and reliable web browsing — multi-tab sessions grouped into named workspaces, biometric lock, host-based ad blocking, and password-manager integration with Android credential providers.

Platform
Android 10+
Target
API 36
License
MIT
  • Multi-tab browsing with independent tabs
  • Named workspaces for organized activity
  • Password-manager integration via Android autofill providers
  • Biometric lock protecting the browser
  • Host-based ad and tracker blocking
  • Session restore without a user-facing sessions collection
KotlinJetpack ComposeAndroid WebViewRoomHilt
Source
FINTECH LEDGER(2024)

Khatawali

Mobile Credit & Debit Ledger

A mobile-first credit/debit ledger built with React, Supabase, and Capacitor Android — Google login, person-wise ledgers with category-level separation, chat-style transaction history, PDF/Excel export, and Android-ready flows like contact import and biometric lock.

Stack
React + Vite
Backend
Supabase
Android
Capacitor
  • Person-wise credit/debit ledgers with live balances
  • Category-separated ledgers for the same person
  • Google OAuth with Android deep-link callback handling
  • Receive / Give quick actions and person transfers
  • Per-person PDF export and Excel data out
  • In-app contact import and optional biometric lock
React 18ViteSupabaseCapacitor
Source

/ About

PROFILE — 02

Model → API → Interface
Designed as one system.

00
Systems shipped end-to-end
0.0000
ROC-AUC on held-out botnet data
<0ms
To process 130K events
00
Federated academic APIs integrated

I'm Pratham — a full-stack and AI developer. I design and ship complete systems: the model, the API, and the interface it deserves. My work spans graph neural networks for threat detection, LLM tooling for databases, computer vision for agriculture, and research infrastructure — each one built end-to-end with production discipline.

That discipline is concrete: typed codebases, read-only execution guards, privacy-preserving schema handling, confidence gating, and latency measured in milliseconds on six-figure event volumes. A system that can't survive contact with real users and real data isn't finished.

I build across the whole path — Python and PyTorch where the math lives, FastAPI where the service lives, React and Next.js where the product lives. Complex technical ideas don't have to feel complex. Making them feel inevitable is the job.

Intelligence in the model is table stakes — the craft is in the system around it.

// Trajectory

  1. 2023Foundations

    Self-taught from documentation, not tutorials — Python, algorithms, and the first neural networks trained on borrowed GPU time.

  2. 2024First system in the field

    Shipped Khatawali — a mobile-first credit/debit ledger that replaced the paper notebook with person-wise, category-separated accounts.

  3. 2025Field products

    ShetVaidya and Tenora — computer vision meeting real agriculture, and offline-first property accounting that runs entirely on the device.

  4. 2026NowLive

    TGDetect, Verbis, Kivo, and Veyla — temporal graph ML for threat detection, LLM tooling that treats privacy as architecture, a research evidence desk, and a private workspace browser. Open to roles, collaborations, and ambitious builds.

/ Capabilities

STACK — 03

The tools I reach for when turning hard problems into shipped products.

Frontend

Interfaces, interaction, and motion

  • React
  • Next.js
  • TypeScript
  • Tailwind CSS
  • JavaScript

Backend

APIs, services, and systems

  • Python
  • FastAPI
  • REST APIs
  • Kotlin
  • VS Code API

AI / ML

Models, graphs, and inference

  • PyTorch
  • GraphSAGE
  • TensorFlow
  • MobileNetV2

Data

Storage, queries, and pipelines

  • PostgreSQL
  • Prisma
  • Supabase
  • Cloudflare R2

Tools

Infrastructure and workflow

  • Docker
  • GitHub Actions
  • Figma
  • Git

Also fluent in

  • Jetpack Compose
  • Capacitor
  • Temporal Graph Learning

— and whatever the problem actually demands.

/ Contact — Have a hard problem?

Let's buildsomethingintelligent

I'm open to roles, collaborations, and ambitious builds in full-stack development, applied AI, and machine learning systems.

prathamwork2511@gmail.comDownload CVPDFgithub.com/Pratham2511