Pfizer Externship via Extern

AI-focused recent graduate building practical experience

I study AI and am actively gaining hands-on skills through projects and a Pfizer externship to make my resume competitive for jobs and internships.

View my work Message Me

Work samples

A selection of practical AI projects plus work from the Pfizer Advanced: AI-Powered Document Insights & Data Extraction externship, showing applied model and data skills.

Kaksha – Satellite Collision Screening
Kaksha – Satellite Collision Screening
Reduced screening time for 75 Indian satellites against 16,760 objects from an estimated 2.5 hours to 74 seconds by computing each orbit once with SGP4 and filtering impossible pairs by altitude. Accelerated on-demand screening of a single satellite from 19 seconds to 0.77 seconds with zero missed close approaches, by adding an altitude filter that skips 94% of impossible object pairs. Delivered a Spring Boot REST API (Spring Data JPA, MySQL) with 36 ms median and 65 ms p95 response time, backed by 72 JUnit tests and deployed on AWS EC2 with Docker Compose and HTTPS.
JavaSpring BootSpring Data JPAMySQLDockerAWS EC2
AI Interview Practice Platform
AI Interview Practice Platform
Built a RAG pipeline that generates tailored interview questions from a candidate’s resume, cutting LLM context by 58% with no loss in retrieval recall (96%) by chunking per accomplishment instead of per section, with FAISS retrieval. Designed an LLM-based resume-bullet scorer that separates strong bullets from generic ones (average 0.84 vs 0.18 across 36 labeled examples) by grading four signals: numbers, named tech, action and scope. Hardened the LLM scorer against prompt injection, cutting the worst-case score of 12 attack payloads from 1.00 to 0.35 by withholding credit from lines flagged as instructions.
PythonFlaskMySQLFAISSLLM APIsRAG
Sign Language Recognition Platform
Sign Language Recognition Platform
Built a real-time sign-language recognition tool that gives learners instant feedback, reaching about 83% accuracy across 290 sign classes, by training a Bidirectional LSTM on 5,700+ hand-landmark sequences. Protected user privacy and kept the backend lightweight by running MediaPipe in the browser and sending only 3,780 landmark values per gesture (30 frames x 126) instead of raw video, normalized with scikit-learn’s RobustScaler. Secured the Flask REST API with JWT authentication and salted password hashing over MySQL, verified it with 48 pytest tests, and deployed it live with Gunicorn, Docker and HTTPS.
PythonFlaskTensorFlowMediaPipeMySQLReactDocker

About me

I study AI and am actively gaining hands-on skills through projects and a Pfizer externship to make my resume competitive for jobs and internships.

I am Shruti Vishwakarma, a recent bachelor-degree graduate focused on Artificial Intelligence. I am building practical experience through projects and an active Pfizer externship to strengthen my resume and prepare for AI roles and internships.

Externships

Pfizer Advanced: AI-Powered Document Insights & Data Extraction Externship

Pfizer

Experience

Software Developer Intern

Toss Solutions · March 2025 – May 2025

Education

Gyan Ganga College of Technology

Bachelor of Technology in Computer Science and Engineering – CGPA: 7.86/10 · Class of 2025

Skills

Artificial IntelligenceAI projectsDocument data extractionModel developmentData processingBackend developer

Back to work

Kaksha – Satellite Collision Screening

Reduced screening time for 75 Indian satellites against 16,760 objects from an estimated 2.5 hours to 74 seconds by computing each orbit once with SGP4 and filtering impossible pairs by altitude. Accelerated on-demand screening of a single satellite from 19 seconds to 0.77 seconds with zero missed close approaches, by adding an altitude filter that skips 94% of impossible object pairs. Delivered a Spring Boot REST API (Spring Data JPA, MySQL) with 36 ms median and 65 ms p95 response time, backed by 72 JUnit tests and deployed on AWS EC2 with Docker Compose and HTTPS.

JavaSpring BootSpring Data JPAMySQLDockerAWS EC2

View all work

Back to work

AI Interview Practice Platform

Built a RAG pipeline that generates tailored interview questions from a candidate’s resume, cutting LLM context by 58% with no loss in retrieval recall (96%) by chunking per accomplishment instead of per section, with FAISS retrieval. Designed an LLM-based resume-bullet scorer that separates strong bullets from generic ones (average 0.84 vs 0.18 across 36 labeled examples) by grading four signals: numbers, named tech, action and scope. Hardened the LLM scorer against prompt injection, cutting the worst-case score of 12 attack payloads from 1.00 to 0.35 by withholding credit from lines flagged as instructions.

PythonFlaskMySQLFAISSLLM APIsRAG

View all work

Back to work

Sign Language Recognition Platform

Built a real-time sign-language recognition tool that gives learners instant feedback, reaching about 83% accuracy across 290 sign classes, by training a Bidirectional LSTM on 5,700+ hand-landmark sequences. Protected user privacy and kept the backend lightweight by running MediaPipe in the browser and sending only 3,780 landmark values per gesture (30 frames x 126) instead of raw video, normalized with scikit-learn’s RobustScaler. Secured the Flask REST API with JWT authentication and salted password hashing over MySQL, verified it with 48 pytest tests, and deployed it live with Gunicorn, Docker and HTTPS.

PythonFlaskTensorFlowMediaPipeMySQLReactDocker

View all work