Akshit Mehta

Bachelor of Technology in Information Technology, CGPA 9.30

Relevant coursework: DSA, DBMS, OOP, Machine Learning, Statistics, Software Engineering

• Working on software virtualization using QEMU and related tooling.
• Automating virtualization workflows with Python scripts that drive QEMU.

• Led Laravel-to-Django migration design, converting legacy routes, controllers, jobs, Eloquent models, and request flows into Django/DRF architecture.
• De-risked OTP and admin auth, Firebase, Razorpay, push notifications, sessions, and password resets through integration risk documentation.
• Built a DRF migration scaffold with user models, serializers, viewsets, JWT patterns, and Faker-based local data validation.

FastAPI, LangGraph, Temporal, PostgreSQL, Next.js

• Built a live company-check agent over records, reviews and news with LangGraph, Temporal and FastAPI; a reviewer can answer its questions, pause it and give feedback while it runs.
• Streamed every step from an append-only PostgreSQL event log over Server-Sent Events into a Next.js UI that is a pure reducer of that log, so any page reload replays the same state.
• Designed one general chart tool (line, bar, area by week to year, up to four series) where the model only describes the series and code computes every figure, so rating trends and company comparisons cannot contain invented numbers.
• Handled similarly named companies, evidence exclusion with automatic chart and finding updates, and Gemini rate limits; verified with 52 backend tests and a 26-check live scenario suite.

Vue 3, Flask, Celery, Redis, ChromaDB

• Built a deployed RAG developer tool that indexes GitHub repositories, source code, pull requests, merge notes, and diff summaries for natural-language codebase Q&A.
• Engineered asynchronous ingestion with Celery, Redis, and ChromaDB; used tree-sitter for AST-aware class and method boundary chunking.
• Implemented Server-Sent Events streaming APIs, reranking, and retry backoff for Gemini/Gemma queries while handling HTTP 429 rate limits.
• Reduced worker RAM usage from 500MB+ to under 150MB for a 512MB VPS and accelerated CI image builds by 6.3x.

FastAPI, PyTorch, Redis Streams, Next.js, PostgreSQL

• Built a continuous authentication system fusing One-Class SVM, LSTM Autoencoder, and TCN Autoencoder models, achieving 91%–93% genuine verification and zero false acceptances in impostor testing.
• Extracted 23-dimensional keystroke timing features and 7-dimensional mouse kinematic features for speed-scale invariant biometric verification.
• Engineered score fusion across keyboard and mouse signals with under 50 ms inference per sequence window and built a FastAPI/Next.js testing sandbox.
• Supported enrollment, verification, identification, and live authentication workflows with SQLite-backed profile storage.