Work & Projects

Systems I've designed, built, and kept running.

Each entry follows the same shape: what was broken, why normal approaches failed, what I actually built, and what happened.

Anaya - Legal AI Document Intelligence

Founder & CEO · September 2025 - Present

What was broken

Law firms manually processing 100+ page contracts. Reviews took days. No systematic extraction of key terms, obligations, or risks.

Why normal automation failed

Generic OCR tools couldn't understand legal context. Simple keyword search missed semantic meaning. No way to track citations or verify claims.

What I built

3-agent pipeline: text extraction → RAG-based context retrieval → summarization with citation tracking. React/Next.js frontend with async state management and real-time browser-based processing. Secure authentication and document sharing via Firebase.

Outcome

Deployed production platform processing legal contracts. Multi-agent orchestration with parallel processing. Built, shipped, and iterated based on real document workflows.

React · Next.js · TypeScript · Firebase · RAG · Multi-agent orchestration

Distributed AI Workflow Orchestration Platform

Solo Engineer · October 2025

What was broken

AI workflows dying mid-execution with no way to recover. When infrastructure failed, entire multi-step processes lost state. No visibility into what step failed or why.

Why normal automation failed

Simple APIs with no durability. Crashes meant starting over from scratch. No checkpointing or replay capability.

What I built

Fault-tolerant workflow engine using Temporal Cloud with pause/resume/replay and checkpoint recovery. Event-driven architecture with Redis queue and PostgreSQL for state persistence. Built observability layer with per-step logging, token tracking, and latency profiling.

Outcome

Handled 50+ concurrent workflow executions. Sub-200ms API response times. Zero state loss during failures. Idempotent REST endpoints for production reliability.

Python · FastAPI · Temporal Cloud · Redis · PostgreSQL · React

Blissm - AI Mental Health Companion

Founder & Lead Engineer · November 2024 - May 2025

What was broken

Mental health apps with no memory. Users had to re-explain their situation every session. No continuity or personalization across conversations.

Why normal automation failed

Stateless LLM APIs that forgot context. No conversation history. Generic responses that didn't adapt to individual users.

What I built

Cross-platform mobile app (iOS/Android) with conversation memory system maintaining context across multi-session interactions. Dynamic prompt engineering and personalization algorithms. Real-time mood tracking with sentiment analysis and emotional pattern recognition.

Outcome

Validated with 10+ professionals (therapists, counselors). Shipped production app with session continuity. Learned mobile development and AI personalization patterns.

React Native · Expo · Firebase · GPT-4 API · Sentiment Analysis

Research & PM Platform for Universities

Project Lead · Academic Project

What was broken

Universities coordinating research across teams using spreadsheets and email. No centralized repository for projects, patents, or team formation.

Why normal automation failed

Existing tools weren't designed for academic workflows. No matching algorithms for connecting students with faculty guides or research teams.

What I built

Multi-tenant SaaS platform connecting students and faculty for research collaboration. Built local university-specific repository for research and IP management. AI-powered recommendation for guide matching, team matching, and startup formation. Led team of 4 through full development lifecycle.

Outcome

Deployed functional platform used within university. Learned team leadership, requirements gathering, and architecture design for multi-tenant systems.

React · Next.js · TypeScript · PostgreSQL · AI Recommendations

NGO Attendance & Analytics System

Software Engineer Intern · November 2023 - April 2024

What was broken

Education NGO tracking thousands of students across multiple centers using paper records. No way to analyze attendance patterns, retention rates, or program effectiveness.

Why normal automation failed

Manual paper-based workflows were error-prone and slow. No centralized data. No analytics or insights.

What I built

Full-stack mobile application digitizing attendance and admission records. Data analytics dashboard with Chart.js visualizations tracking attendance patterns, student retention, and program metrics.

Outcome

Reduced administrative overhead by 60% through automation. Improved data accuracy across all study centers. Learned to build for non-technical users with minimal training.

React · Node.js · MongoDB · Chart.js · Mobile-first Design

Kiki - Event Discovery App (5-Day Prototype)

Solo Developer · October 2025

What was broken

Finding local events meant checking dozens of fragmented sources manually. No unified discovery experience.

What I built

Full-stack event discovery app shipped in 5 days. Firebase backend, Maps API integration, agentic AI chain for recommendations. Authentication, real-time data, and serverless functions.

Outcome

Concept to deployed production in 5 days. Proved ability to move extremely fast from idea to functional product without sacrificing quality.

Next.js · Firebase · Google Maps API · AI Agents · Vercel

Neural Network Classifiers for PolSAR Satellite Imagery

Research Assistant · Research Lab, VIT Mumbai

What was broken

Traditional classifiers struggled with Polar Synthetic Aperture Radar (PolSAR) satellite images for environmental monitoring. Poor accuracy on mangrove detection tasks.

Why normal automation failed

Classical ML approaches couldn't handle high-dimensional decomposed PolSAR data. No systematic comparison of optimization functions.

What I built

Neural network classifiers for mangrove detection using decomposed PolSAR images. Comparative analysis of optimization functions (Adam, SGD, NN) evaluating performance impact on classification accuracy. Collaborated on experimental design, model training, and statistical analysis.

Outcome

Co-authored 2 papers published at IEEE IGARSS 2024 (Athens, Greece) — premier conference in geoscience and remote sensing. Learned rigorous research methodology and academic publishing.

Python · TensorFlow · Neural Networks · PolSAR · Remote Sensing

Have a system that needs to work reliably?

Email me