Vasishta Nandipati Open to internships Resume

AI/ML engineer · CSE (AI & ML), Manipal Institute of Technology · Class of 2027

Vasishta Nandipati

I build machine-learning systems, data pipelines and security tooling in Python. My work runs from an IEEE-published path-planning method to leakscan, a secret scanner with 3,800+ downloads on PyPI.

Open to internships for my Practice School semester, 2027

AI/ML engineering Software engineering Data Research

View my work Download resume Let’s talk
Fig. 1 A* over a slope-penalised terrain cost map, a simplified version of the method in my IEEE paper. Click the map to set a new goal.

About

I work on the part of machine learning that has to run outside a notebook: pipelines that hold up on messy data, tools other developers install, and algorithms that have to work on real terrain.

What I work on

  • Applied ML and computer vision. Depth-image path planning with NumPy and OpenCV; models in PyTorch, TensorFlow and scikit-learn.
  • Data workflows. Reproducible pipelines for large healthcare datasets at Mangomelon AI.
  • Security tooling. leakscan, plus the Google Cybersecurity certificate behind it.
  • Product. Founding member at Velosta, a B2B SaaS for rental operators.

How I build

  1. Ship where people install things. leakscan is on PyPI, Homebrew and the VS Code Marketplace, not just GitHub.
  2. Test the boring parts. 117 tests and a CI matrix came before the newest features.
  3. Audit my own code. I found and fixed a ReDoS in leakscan’s ignore-file parser and locked its config file to 0600.
  4. Keep data local when I can. DataForge runs its LLM through Ollama on the user’s machine.

Experience

  1. 2026May – Jul

    Digi edZe Soft Solutions

    Software Engineering Intern

    Hyderabad

    • Worked on Chat4ED, an agentic conversational BI platform built on a five-agent anti-hallucination pipeline with sovereign on-premise deployment.
    • Built the end-to-end outbound infrastructure and GTM content library, driving a 300+ prospect pipeline across verticals toward paid enterprise PoC agreements.

    Agentic BIApollo.ioSales NavigatorInstantly.ai

  2. 2026Jan – now

    Mangomelon AI LLC

    Data Analyst

    Remote

    • Designed AI-driven analytical workflows that automate data cleaning, feature extraction, statistical evaluation and pattern discovery.
    • Built reproducible pipelines for ingesting, validating, transforming and visualizing large structured healthcare datasets.

    Healthcare dataPipelinesStatistical analysis

  3. 2025Oct – now

    Velosta Technologies LLP

    Founding Member, Product

    Hyderabad

    • Co-founded Velosta Vendor, a B2B SaaS platform for bike and car rental operators in India, covering Aadhaar and DigiLocker KYC, WhatsApp automation, GST billing and multi-branch management.

    B2B SaaSKYCProduct

  4. 2025May – Jul

    Algonomy

    Summer Intern

    Bangalore

    • Implemented optimized logic for the Production Suffix module, improving deployment-pipeline assembly.
    • Built dataset ingestion utilities, CustomTkinter GUI components and Python REST interfaces for internal engineering workflows.

    PythonREST APIsCustomTkinter

Selected projects

Open source · Python · v0.4.0

leakscan

A pure-Python CLI that finds leaked API keys and secrets in source trees, staged changes, git history and public GitHub profiles, and checks whether they still work.

Problem
Secrets get committed, deleted and pushed, but they stay in git history. The common scanners need Docker, a config file or a Go binary before the first scan.
What I built
107 detection patterns across severity tiers, Shannon-entropy scoring, history and staged-diff scanning, read-only live verification for 32 services, SARIF output and a built-in pre-commit hook. Each finding explains its blast radius, where to rotate the key, and who committed it.
Engineering decision
--staged scans only the diff about to be committed, so the pre-commit hook stays near-instant on large repos. Full scans are multi-threaded with chunked reads to keep memory flat.
Evidence
3,800+ downloads, 117 tests across Python 3.10 to 3.12 in GitHub Actions, and installs via pip, pipx, uv, Homebrew and a VS Code extension.
$ pip install leakscan
$ leakscan scan . --history --verify

[CRITICAL] AWS Access Key ID
  File: config.py:3
  Match: AKIAIOSFODNN7EXAMPLE
  Blast radius: Full cloud account compromise
  Rotate at: console.aws.amazon.com/iam
  Git blame: commit a1b2c3d (45d ago, main)
  Secret age: 45d scraping risk: CRITICAL
Output format from the project README, using AWS’s documented example key.
DataForge desktop app: a domain keyword field, rows-per-dataset and variation settings, and start and stop generation controls.
The DataForge desktop client.

Python · Ollama · Flask · CustomTkinter

DataForge

A synthetic-data generator for ML prototyping when the right dataset doesn’t exist or can’t be shared.

What I built
Automated Kaggle dataset discovery for a domain keyword, then a locally hosted LLM (Mistral via Ollama) generates synthetic variations at a chosen size. Exposed through a desktop GUI and a REST API.
Engineering decision
Running the model locally means reference data and prompts never leave the machine. The REST endpoints are key-gated, with dataset listing, generation and zipped downloads.
Source on GitHub

Python · SerpAPI · BeautifulSoup · fpdf2

Vendor Due Diligence

Automated vendor screening for procurement: type a company name, get a compliance-ready evidence report.

Aggregates web search results, scans each source for risk signals and archives the evidence into a PDF report. A background worker thread feeds the GUI through a queue so the interface stays responsive, and failed page fetches are retried automatically.

Source on GitHub
  1. 01Company name
  2. 02Web search via SerpAPI
  3. 03Fetch and parse sources
  4. 04Risk-term screening
  5. 05PDF evidence report

Also on GitHub

Research

IEEE · MoSICom 2025 · Dubai

Enhanced A* Path Planning with Potential Fields for Terrain-Aware Robot Navigation

K. Davasam, A. Dittakavi, V. Nandipati, J. James, R. Mahadeva, K. Chadaga · Manipal Institute of Technology

Abstract, in brief

Standard A* minimises path length and ignores the ground. We combine A* with a potential field built from RGB and depth images, so the planner also weighs slope, depth discontinuities and obstacles. The result is a route that is short, but also safe and feasible on rugged terrain.

Contribution

A global A* search guided by a terrain-derived cost field. The potential field supplies local obstacle repulsion, and the global search gives it a way out of the local minima that pure potential-field methods get stuck in.

Python · NumPy · OpenCV · Gazebo

Read on IEEE Xplore

Method

  1. 01Depth normalisation Raw depth scaled to [0, 1] for consistency across resolutions.
  2. 02Gradient Slope from depth gradient magnitude.
  3. 03Potential field Gaussian-smoothed obstacle, slope and goal-attraction fields.
  4. 04Modified A* Priority-queue search over terrain-weighted costs, then drawn over the RGB frame.

Cost = Base × (1 + 20S + 10(1 − D) + 5P + 15DP)

S slope · D depth · P potential field · DP depth change. Slopes above 0.7 take a further 3× penalty.

Capabilities

Grouped by what I use them for, with where each one shows up.

AreaToolsWhere it shows up
AI / ML PyTorch, TensorFlow, Keras, scikit-learn, OpenCV IEEE paper, DataForge, disease predictor
Data Pandas, NumPy, SQL, Tableau Mangomelon AI, Google Advanced Data Analytics
Engineering Python, Java, C, JavaScript, REST APIs, Docker, Linux, Git leakscan, Algonomy, CUDA benchmarks
Security Secret detection, IDS/IPS, OWASP principles, SIEM tools leakscan, Google Cybersecurity certificate

Education

2023 – 2027

B.Tech, Computer Science & Engineering (AI & ML)

Manipal Institute of Technology, Manipal

Coursework includes machine learning, deep learning, operating systems, computer networks and data structures.

Certifications

  • Google Advanced Data Analytics Certificate2026
  • SEBI Investor Certification2025
  • Google Cybersecurity Professional Certificate2024

Available for Practice School 2027

Hiring an AI/ML or software engineering intern?

I’m looking for an internship for my Practice School semester, in AI/ML engineering, software engineering, data or research. Email is the fastest way to reach me.

vasishtavj@gmail.com