I build AI agents and the software around them. I’m finishing a master’s in Computer and Information Science at Penn, and this summer I was an AI engineering intern at Postman.
I’ve shipped work at Postman Ericsson
I like problems that start out vague. At Postman that meant recruiting pipelines run by LLM agents, and a tool that ships a Postman Flow to whatever backend fits it. At Ericsson it meant more than a hundred features for 5G revenue management. Along the way I’ve published twelve papers, mostly on image retrieval and machine learning for healthcare.
AI Engineering Intern
Software Engineer
Software Engineer Intern
Research Intern
MSE, Computer and Information Science
Coursework in operating systems, machine learning, and computer vision.
B.Tech, Computer Science and Engineering
Coursework in data structures, image processing, and artificial intelligence.
My venture
A recruiting engine
It reads a job description, scores candidates against the priorities you set, and writes LinkedIn x-ray searches to find more of them. It connects to the Manatal ATS and exports to CSV, PDF, and JSON.
Built at Postman
Build a flow once, ship it anywhere
Build the logic in Postman Flows, or describe it to an LLM over MCP. Portage reads the flow, works out what it needs, generates the files each backend expects, and ships it to Postman, Astro, or n8n. The flow never changes, so you can let Portage pick the destination or choose one yourself.
Not public yet
Ranks a role’s applicants, or searches a whole candidate database. Every candidate is embedded into a persistent Qdrant index; the closest matches are then scored by an LLM with weights that adapt to each job description. Email and scheduling happen in the same window.
A clinical web app for studying cerebral blood flow in patients with an LVAD heart pump. From raw signals sampled at 125 Hz it computes autoregulation (CA and Mx), cerebrovascular reactivity, and per-beat pulsatility, and separates heart-driven beats from pump-driven ones. It is a full Python and Flask port of a lab’s MATLAB tool, with Excel and JSON export.
Captions your photo, writes a scene with an LLM, renders the frame with Stable Diffusion, and swaps your face in, all behind a Gradio interface. Built for Penn’s computer vision course, CIS 5810.
A Streamlit chatbot that answers free-form HR questions from live Workday data and company policy. An LLM agent picks from read-only tools that query Workday over OAuth2 REST, and you can switch models from the sidebar. Built internally at Postman, so the code is private.
Employers say exactly what evidence they accept for a role, candidates submit real work, and an LLM produces a ProofCard: a pass or fail verdict tied to that evidence rather than to a resume.
A file system built on N-ary trees and hashing, with a JSON hash tree for change detection. Parallelized with OpenMP, it improved backup performance by more than 100 percent over traditional systems.
A hybrid feature extraction pipeline combining DWT, PCA, GLCM, and OKE, tested on the BRATS 2017 brain tumor dataset. It improved retrieval accuracy by up to 28 percent over baseline models. Published in AIP Conference Proceedings 2916 (2023).
A deep learning model on an Inception v3 backbone that grades diabetic retinopathy from retinal scans, with a Flask front end so clinicians can use it.
Twelve publications between 2021 and 2025, on machine learning, computer vision, healthcare, and security.
I’m glad to talk about agents, full stack products, ML research, or a role where I’d get to build them.
nitinlodha2812@gmail.com