Dublin, Ireland · MSc Artificial Intelligence, National College of Ireland

From genomes
to agents.

I'm Aditya Sharma, an AI engineer in Dublin. I build LLM agents, retrieval systems and data pipelines. I started in bioinformatics, analysing around 10,000 cancer mutations at IIT Bombay, so I test my models the way I was taught to test a hypothesis.

See my work Get in touch
5th
of 7,700+ participants at the IIT Hyderabad National AI/ML Hackathon 2026
~10k
TP53 mutations analysed for my thesis research at IIT Bombay
14mo
full-time developer experience across two companies
3
live AI products you can try on this page

NowStudying for an MSc in AI and building a galaxy classifier that has to hold up under distribution shift.

01Work

Projects

Things I have built, most with public code. Three have live demos you can open right now.

Hackathon · team of 3 · 2026

IntelliCredit

An AI credit-analysis pipeline for Indian small businesses. It reads annual reports, bank statements and GST filings and turns them into a credit health dashboard.

  • Parses PDF, CSV, Excel and DOCX with pdfplumber and PyMuPDF, with Tesseract OCR for scanned pages.
  • Claude extracts metrics as structured JSON; a regex analyser and derived-metrics engine take over when the API is unavailable.
  • XGBoost credit scoring with SHAP explanations; handles lakh/crore units and bracketed negatives.
  • Python
  • Claude API
  • XGBoost
  • SHAP
  • Tesseract
  • Flask
  • TypeScript
Code

Hackathon build · 2026

Interview Buddy

An AI mock interviewer you talk to on camera. It researches how the role is interviewed, asks questions by voice and reports on your delivery.

  • Tracks eye contact and smiling with MediaPipe Face Landmarker, plus pace and filler words.
  • A per-role dashboard charts progress across sessions.
  • FastAPI
  • Claude API
  • MediaPipe
  • Web Speech API
  • Chart.js
Live demo Code

Solo project · 2026

Repositioning Intelligence

A research platform for drug repurposing that links drugs, diseases, molecular targets and evidence records in one place.

  • AI summaries per drug–disease pair are reviewed by a person and graded, with citations from PubMed, ClinicalTrials.gov and FDA labels.
  • Confidence scoring and a reviewer login with an audit trail. Built for generating hypotheses, not for patient care.
  • Next.js
  • TypeScript
  • React
  • Prisma
  • Tailwind
  • Vitest
  • Playwright
Live demo Code
Histogram of relative fitness scores for p53 variants, showing two peaks
Thesis graded O (10/10)

Research · IIT Bombay · 2025

The fitness landscape of p53 mutations

My BTech thesis, carried out at IIT Bombay under Prof. Supreet Saini. I analysed how ~10,000 mutations in TP53, the most frequently mutated gene in human cancer, change cancer-cell fitness.

  • Gaussian mixture models show a two-peaked fitness landscape; KS, Mann–Whitney U and Cohen's d compare mutation classes.
  • Some "silent" synonymous mutations still change fitness; a sliding-window codon adaptation index maps fast and slow translating regions.
  • Exhaustive epistasis scan of 797 mutation pairs: 535 negative, 160 none, 52 sign and 50 positive.
  • Reproducible: GitHub Actions re-runs every notebook on each push, and each one opens in Colab.
  • Python
  • pandas
  • NumPy
  • SciPy
  • scikit-learn
  • GitHub Actions
Code Thesis (PDF)

Prototype · 2026

carefully

A multilingual, voice-first care-navigation prototype. It helps people put what they are feeling into words and prepare to talk to a clinician. It does not diagnose.

  • Nothing goes to the AI model without explicit consent; a deterministic emergency-phrase check runs independently of it.
  • A Vercel serverless function calls Cerebras inference, keeping the API key off the browser.
  • JavaScript
  • Web Speech API
  • Vercel Functions
  • Cerebras
Live demo Code

Pipeline · 2026

Lead intelligence pipeline

A daily pipeline that finds people unhappy with a competitor's product, ranks them as leads and hands a digest to a human for review.

  • Collects from web research, Reddit and RSS, removes duplicates by URL, content hash and fuzzy match, then enriches and scores with OpenAI.
  • Slack digests and Excel reports; FastAPI endpoints, health checks, Docker and CI on GitHub Actions.
  • Python
  • FastAPI
  • PostgreSQL
  • OpenAI API
  • Docker
  • pytest
Code

More I have built

  • DRAP: Dublin Rental Application Processor RPA plus a locally hosted LLaMA 3 agent that validates documents and ranks applicants, with a bias audit and a GDPR consent and deletion ledger. MSc module project · LLaMA 3 · SQLite · Streamlit
  • Ghost AI A local-first voice agent for Google Meet and Teams that hears its user's name and answers within five seconds. Whisper · Llama 3.3 70B on Groq · edge-tts
  • Real-time BTC prediction Live market data from the Binance WebSocket API streamed through Kafka to a live prediction dashboard. Team project · Kafka · Airflow · Docker Compose
  • Air-quality forecasting (PM2.5) Compared statistical, ML and deep-learning models; gradient boosting with pollutant features reached R² 0.9953, ahead of ARIMA, LSTM and AutoML. Time series · gradient boosting · LSTM
  • Worker position and safety monitor Indoor tracking from phone sensors with a Kalman filter, plus YOLOv8 floor-safety monitoring on a webcam feed. YOLOv8 · OpenCV · Flask
  • KramaAI marketing site A static marketing site for a business management product: pricing, FAQ, demo booking and legal pages. HTML · CSS · JavaScript · Vercel

02Experience

Where I have worked

  1. Mar 2026 – Aug 2026Remote

    Developer · Code Ninjas HQ

    Joined as a junior developer reporting to the CTO, working on an AI-powered business management system for multi-location franchises.

    • Built an all-in-one chatbot in Django that adapts to any business and answers from that business's own data, using MCP servers and vector databases.
    • Built chatbot and agent workflows (intent classification, entity extraction, conversation context, API routing, fallbacks) and RAG with semantic retrieval and reranking.
    • Built the React and Node.js website that sells the product, including the purchase-to-onboarding flow, plus data pipelines for event discovery and lead scoring.
    • Django
    • MCP
    • RAG
    • Vector databases
    • React
    • Node.js
    • Azure
  2. Jul 2025 – Feb 2026Mumbai

    Developer · Ambarnath Transformers & Electricals

    Worked directly with the CEO, Director and owner of an electrical transformer manufacturer.

    • Built an analytics dashboard that combines Python web scraping with ML classification to surface relevant public tenders.
    • Produced weekly Power BI insight reports for senior staff from analysis of internal company data.
    • Built the company's first website end to end and set up its hosting, releases and CI/CD.
    • Python
    • Web scraping
    • Machine learning
    • Power BI
    • JavaScript
    • CI/CD
  3. Dec 2024 – Jun 2025Mumbai

    Data Analysis & Research Intern · IIT Bombay

    Statistical analysis of a public p53 mutation library (GEO GSE115072) for my BTech thesis, supervised by Prof. Supreet Saini, Department of Chemical Engineering. See the project.

    • Python
    • pandas
    • Statistical modelling
    • Genomics
  4. Jul 2023 – Dec 2024Navi Mumbai

    Data Science Intern · D.Y. Patil University

    • Research under the Head of Bioinformatics on nutrient-based antiviral molecules against flaviviruses, using molecular docking and literature analysis; presented a poster at ICBSSF, an international conference.
    • Built psoriasis image classifiers on 1,000+ labelled images and assisted with deep-learning research and manuscript preparation.
    • Deep learning
    • Molecular docking
    • Computational biology
  5. Dec 2023 – Jan 2024Remote

    Data Science Intern · NullClass

    Built end-to-end NLP pipelines for emotion classification from text and voice, and benchmarked LSTMs against transformer models.

    • NLP
    • LSTM
    • Transformers

03Toolkit

What I work with

AI & LLMs

  • AI agents
  • RAG
  • MCP servers
  • Vector databases
  • Reranking
  • Prompt engineering
  • Structured outputs
  • Claude API
  • OpenAI API
  • Llama 3
  • Whisper
  • LLM fine-tuning

Machine learning

  • scikit-learn
  • PyTorch
  • XGBoost
  • SHAP
  • CNNs
  • LSTMs & transformers
  • YOLOv8
  • OpenCV
  • NLP
  • Time series

Data & statistics

  • pandas
  • NumPy
  • SciPy
  • Hypothesis testing
  • Gaussian mixtures
  • ETL pipelines
  • Web scraping
  • Kafka
  • Airflow
  • Power BI
  • Snowflake

Back end & web

  • Python
  • Django
  • FastAPI
  • Flask
  • REST APIs
  • Node.js
  • React
  • Next.js
  • TypeScript
  • Tailwind CSS
  • Streamlit

Data stores & cloud

  • PostgreSQL
  • MongoDB
  • SQLite
  • Prisma
  • SQL
  • Azure
  • Vercel
  • Docker
  • GitHub Actions
  • CI/CD

Bioinformatics

  • Genomic mutation analysis
  • Molecular docking
  • Computational biology
  • PubMed & ClinicalTrials.gov
  • R
  • Java & BioJava

04About

A scientist's habits, an engineer's job.

I grew up in a family of doctors and studied bioinformatics and data science, which meant learning to distrust a result until the statistics backed it. My thesis at IIT Bombay was months of exactly that: thousands of mutations, careful tests, and figures that had to survive questions.

Since then I have mostly been building: chatbots and agents that answer from a company's own data, pipelines that clean and rank messy web data, and small products people can try. I like work where the AI has to be useful and checkable at the same time.

I am now doing an MSc in Artificial Intelligence at the National College of Ireland, including a module on data governance, ethics and sustainability. Outside code, you will find me at the gym six days a week.

Education

2026 – 2027 · in progress

MSc Artificial Intelligence

National College of Ireland, Dublin

Machine Learning · Principles of AI · Programming for AI · Data Governance, Ethics and Sustainability

2021 – 2025

BTech Bioinformatics and Data Science

D.Y. Patil Deemed to be University, Navi Mumbai

Thesis carried out at IIT Bombay, graded O (10/10)

05Contact

Let's build something useful.

I'm open to graduate and junior roles in AI engineering, data and software, and to research collaborations. Email is the quickest way to reach me.

Email me LinkedIn GitHub