← Back to the homepagePROJECT INDEX / PRIYANSH SHARMA

SELECTED ENGINEERING & INDEPENDENT PROJECTS

Built at work. Built on my own.

Some problems came with real production constraints. Others began because I wanted to figure something out. Both have shaped how I design, build, and think about software.

THIS COLLECTION
03Professional engineering stories
01Independent product in focus
07Earlier technical projects

01 / PROFESSIONAL ENGINEERING

Systems with real constraints.

Work on complex backend systems, where architecture has to account for ownership, integration boundaries, failure, and recovery—not just the happy path.

01 / DISTRIBUTED PROCESSINGTHALES

Making distributed billing finish correctly.

I architected and implemented the core of a distributed bill-run engine. The challenge was to preserve pricing semantics across parallel stages and keep progress recoverable when workers were interrupted.

JavaCassandraConcurrencyRecovery
Read engineering case study
A SYSTEM IN STAGES FIG. 01
MAPREDUCEFINALIZE

Durable coordination between stages

02 / SYSTEM DESIGNTHALES

Making Fleet interactions easier to reason about.

I designed and implemented a framework within an existing order scheduler, separating interaction planning from execution. The work had to respect established behavior while making future changes more predictable and testable.

OrchestrationArchitectureReliability
Read engineering case study
SEPARATE THE CONCERNS FIG. 02
PLANEXECUTE

Decisions and side effects have different responsibilities

03 / SELF-INITIATED ENGINEERINGTEAM ADOPTION

Helping AI understand the code around it.

While working with large backend codebases, I kept seeing the same problem: AI tools lost important architecture, conventions, and context between sessions. I took the initiative to build a persistent engineering knowledge system—not as an assigned project, but because I wanted to fix that problem.

It uses structured architectural context, AST-aware indexing, Qdrant/Neo4j retrieval, and MCP integrations. What started as my own initiative was eventually adopted by my entire team.

Code intelligenceAST-aware indexingMCPQdrant / Neo4j

Self-initiated during professional engineering work. Described conceptually; no private code or internal architecture is published.

CONTEXT THAT LASTS FIG. 03
01 Understand code structure
02 Preserve engineering knowledge
03 Retrieve context when needed

Built independently · Adopted by the team

These professional engineering stories describe my responsibilities and lessons through conceptual explanations. They do not reproduce employer source code, operational details, or confidential systems.

02 / INDEPENDENT PRODUCT

An idea I kept building.

A project I chose to work on independently, bringing together backend engineering, product design, and applied AI.

FEATURED INDEPENDENT PROJECTAI / PRODUCT ENGINEERING

Job Find Helper — AI Career Intelligence.

Job openings are scattered across company portals, while figuring out which roles actually fit a candidate takes its own effort. I built a platform to collect, synchronize, deduplicate, and classify openings, then rank them using more than keyword matching.

The recommendation approach combines semantic embeddings, skills, and recruitment-history signals with LLM-based explanations of possible skill gaps. Ranking quality is evaluated with Precision@K and NDCG@K.

PythonFastAPIReactEmbeddingsLLMsRanking evaluation
View GitHub repository

The latest implementation described here is private. The linked public repository contains an earlier iteration and will be updated separately.

01 / GATHEROne place for openings

Aggregation, synchronization, deduplication and role classification.

02 / MATCHGo beyond keywords

Semantic similarity, skills and hybrid recommendation signals.

03 / EXPLAINMake rankings useful

LLM-based skill-gap explanations and relevance evaluation.

03 / EARLIER PROJECTS

Where some of the curiosity started.

Before I worked on production systems, I explored computer vision, algorithms, machine learning, and full-stack development. These projects are part of that foundation.

01 / 07COMPUTER VISION · 3D GEOMETRY

3D reconstruction from multiple 2D images

A geometric reconstruction project using front, side, and top views to recognize shapes and generate a 3D SCAD model and point-cloud representation. It taught me how much reasoning sits between recognizing an object and rebuilding it.

02 / 07INFORMATION HIDING · ALGORITHMS

Steganography tools

Experimented with hiding and recovering messages in images, text, audio, and video using LSB encoding, zero-width characters, and other techniques. An exploration of representations and algorithms, not a modern cryptographic security product.

PythonLSB encodingSteganography
03 / 07DEEP LEARNING · COMPUTER VISION

Ultrasound image classification with CNNs

Built a convolutional neural network to classify a breast-ultrasound dataset into benign, malignant, and normal categories. An educational exploration of model architecture, training, validation, and image classification—not a clinical diagnostic system.

PythonCNNImage classification
04 / 07SEARCH ALGORITHMS · GAME AI

Mancala with Minimax

Implemented a two-player Mancala game with player-versus-player and player-versus-bot modes. The bot uses Minimax with alpha-beta pruning to explore moves more efficiently.

PythonMinimaxAlpha-beta pruning
05 / 07EARLY MACHINE LEARNING · WEB APPLICATION

COVID-19 prediction prototype

An early learning project combining a Random Forest classifier with a Flask web application deployed on Azure. It explored how a trained model could be exposed through an application; it was not medically validated.

PythonRandom ForestFlaskAzure
06 / 07DATA STRUCTURES · DATASET OPERATIONS

Trie-based public call operations

A C++ data-structure project exploring trie-based search and operations on a public-call dataset, including phonebook-style lookup and organization of records.

C++TriesData structures
07 / 07FULL-STACK WEB DEVELOPMENT

Shopping application with the MERN stack

A learning project for building an online shopping application with product pages, user flows, a cart, and a MongoDB-backed server. An early end-to-end web development project.

MongoDBExpressReactNode.js

STILL BUILDING

Not a finished list. A record of the work.

I'll keep improving these projects and building new ones. This page is a snapshot of what I've learned so far.