Subject: Ansh Sharma

Code: P-33

Log Date: --.--.----

Observer: Self

APPROVED FOR REVIEW

fingerprint Professional Persona

Subject currently works across data engineering, machine learning, and ML systems. This dossier documents selected projects, experiments, and technical work across data, software, and machine learning, ranging from broader systems work to smaller focused experiments. Individual records cover the problem, implementation, decisions, and results where applicable. The following sections provide a closer look at the work behind the file, including a few unnecessary detours along the way.

Current Status

  • Role:Data Analyst Intern
  • Target:Data Science | ML Engineer
  • Type: Internal Use Only
  • Location: New Delhi, IN
  • Clearance: Level 3

Subject's Whereabouts * linkedin

* Github

* Mail

Service Record
Data Analyst Intern August 2026 - PRESENT
iTech Mission Pvt. Ltd.
  • target Validate government-data modules and trace inconsistencies across source files, frontend inputs, and backend processing.
  • target Prepare and verify master files, indicator metadata, and supporting datasets for ingestion into internal systems.
  • target Perform data cleanup, format checks, and source verification using Excel, SQL, and Python where applicable.
  • target Investigate data-quality issues and recommend validation or preprocessing checks before release.

Experimental Overview (Projects)

EXP_01

Spore

Flagship

Docs ↗
*Currently exceeding 590 Docker pulls.

A local-first data platform for privacy- and cost-conscious users. Built independently by the subject to explore a more self-contained approach to data analysis and ML workflows. Threat model: realistic. Ambition level: concerning.

Spore Diagram
EVIDENCE:Spore served as the primary workspace for the XLM-V Tuktuk HIEN LID project — from data scraping and processing through labeling, model training, evaluation, and result preparation.
EXP_02

XLM-V Tuktuk LID

Side exp

Docs ↗

Named after subject's cat Tuktuk. XLM-V tuktuk HIEN LID is a token-level Language Identification Model specialized for code-mixed Hinglish text.

ExampleEN
textEN
haiHI
actualEN
datasetEN
seHI
nhiHI
Validation: · Achieved 91% weighted F1 and 87% Macro F1 on external LINCE Benchmark, within 1% of best performing model.
XLM-V Hugging Face NLP Data Collection
EXP_03

Wingman

Legacy

Local LLM Integration Environment.

VS Code extension for AI-assisted development without sending your codebase to a third party. Runs entirely on local LLM instances.

VS Code Extension Ollama Local AI
Github ↗

menu_book Academia

Data Annotation and Collection Project

Contributed in a academic project on code-mixed Hinglish text classification. Work involved multi-platform data collection across Reddit, Twitter, Instagram, Facebook, YouTube, and Telegram — each with its own access constraints and workarounds.

Also handled corpus annotation and proposed a confidence-threshold based annotation pipeline to reduce manual bottleneck on a ~25k record dataset.

build Technical Core

  • Languages: Python, R, SQL, NoSQL
  • Databases: PostgreSQL, MySQL, MongoDB, Supabase
  • Data: Pandas, NumPy, Apache Arrow, Apache Parquet, DuckDB
  • AI / ML: Scikit-learn, NLTK
  • LLM Tooling: LangChain, Hugging Face Transformers, Ollama, RAG, Vector Embeddings
  • Infrastructure: Docker, Redis, Apache Superset, Flask, Git