Work Lab
About Resume
Get in touch
GitHub LinkedIn

Ankita Gaikwad · Pune, India

Software engineer.
Curious across the stack.

Backend and cloud are where I've done my best work: two years of production APIs on AWS for healthcare. Full-stack apps and RAG pipelines are where curiosity took me next.

REST APIs Serverless AWS Full-Stack RAG Pipelines Healthcare Tech
View projects Say hello
ankita@portfolio: ~

Welcome. Type help to see what this does, or joke if you need a break.

ankita@portfolio:~$
Scroll

About

A little about me

I like finding out why things break, then turning the fix into something nobody has to think about again.

At Accenture, I worked on the backend of a healthcare platform, and a lot of that time went into failures: tracing where transactions broke, then automating the recovery so the team didn't have to do it by hand.

Outside work, I pick up whatever I want to understand properly. That's how I ended up building a RAG app that cites its sources, and a full-stack journaling app from login to database.

Right now I'm looking for a full-time SDE role. I'm also open to freelance projects and collaborations, so if you have something that needs building, say hello.

Certified

Azure AI Fundamentals (AI-900)

Focus

Backend · cloud · full-stack · RAG

Experience

2+ years in production

Status

Open to SDE roles & freelance

Experience

Where I've worked & studied

Work Jun 2024 — Jul 2026

Associate Software Engineer — Accenture

Backend engineer building serverless microservices that link a vendor system to a Salesforce healthcare CRM: AWS Lambda, Python, Node.js, REST APIs.

200K+Transactions recovered
60%Less manual reprocessing
Days → minFailure detection
  • Architected an automated, idempotent failure-detection and retry system in Python that recovered 200,000+ failed benefit-verification transactions and cut manual reprocessing by 60%.
  • Shipped 23 REST API endpoints on AWS Lambda + API Gateway, handling up to 15K requests a day.
  • Traced repeated failures to Lambda throttling and added CloudWatch alarms on every endpoint, so failures surface in minutes instead of after days of cross-team escalation.
Education 2020 — 2024

B.E., Computer Engineering — PCCOER, Pune

CGPA 8.5 / 10

Projects

Things I've built

Featured · RAG · AI
5 min → 10 s query time

Askive — Chat with your documents

Upload a PDF, DOCX or TXT file and ask it questions. Every answer cites the exact passages it came from. Version 2 adds accounts, hybrid search, caching and a Kubernetes deployment, load-tested with Locust.

Goal
Chat with your own documents, and be able to trust the answers.
Built
Hybrid search (embeddings + BM25, merged with Reciprocal Rank Fusion) over ChromaDB; Gemini answers only from the retrieved passages. JWT accounts, Redis caching, Postgres, Docker and Kubernetes with autoscaling.
Hurdle
Queries took 5+ minutes, uploads hit embedding rate limits, and a load test showed one pod failing 11.7% of requests at 100 users.
Fix
Moved the blocking LLM call off FastAPI's event loop (under 10 s per query) and batched embeddings. Fixed three bottlenecks the load test found; autoscaled to 3 pods: 0.1% errors at 100 users.

Python · FastAPI · ChromaDB · Voyage AI · Gemini · Redis · Postgres · Docker · Kubernetes

View code
Full-Stack

Lantern — Journaling App

A private journal with magic-link sign-in, rich-text entries, images, tags, favorites, search and a calendar view. Row Level Security keeps each user's entries locked to them at the database level.

Next.js · TypeScript · Supabase · PostgreSQL

View code
Creative Coding

Algo-Sonic

A sorting visualizer where every compare and swap plays a note, so you can hear O(n²) against O(n log n). Each algorithm is a Python generator, kept separate from the drawing and sound code.

Python · Pygame · NumPy

View code Play it in the Lab
8 algorithms you can hear: 6 sorts, 2 searches
Backend · Fintech
0 payments lost or paid twice, through worker crashes, network outages and lost replies

Settlement Engine — A payment engine that never pays twice

A bank-style engine that sends payments to simulated Faster Payments and SEPA networks, and keeps going when things break. A live dashboard plays a 90-second chaos demo on the real system.

Goal
Never lose a payment, never pay one twice, and prove it.
Built
FastAPI + PostgreSQL as the source of truth, a Redis queue with atomic Lua claims and leases, and payments idempotent on the ISO 20022 UETR.
Hurdle
A reply can get lost after the money has already moved, so a retry could pay twice.
Fix
Resend with the same UETR, which the network deduplicates. An append-only audit trail and end-of-day reconciliation prove it; 178 tests run in GitHub Actions on every push.

Python · FastAPI · PostgreSQL · Redis · Docker · Prometheus · Grafana · GitHub Actions

View code

Skills

What I work with

Core stack
Python REST APIs FastAPI AWS Lambda PostgreSQL TypeScript React / Next.js RAG

Backend & APIs

Python FastAPI REST APIs Node.js Express.js Java SQL Postman

Cloud & DevOps

AWS Lambda API Gateway CloudFormation CloudWatch CodePipeline Microsoft Azure (AI-900) Docker Kubernetes CI/CD GitHub Actions Prometheus Grafana Infrastructure as Code Git / GitHub Agile/Scrum

Frontend

TypeScript JavaScript React Next.js Tailwind CSS HTML5 / CSS3

AI & Data

RAG Pipelines Vector Search Semantic Chunking Embeddings (Voyage AI) Gemini API HuggingFace Transformers TensorFlow PostgreSQL Redis Supabase ChromaDB

see you soon

Contact

Hiring, or need something
built?

Full-time role, freelance project, or a question about something I've built. My inbox is open.

Email me View résumé
GitHub LinkedIn

I reply to every message. Unread badges stress me out