Nikhil Sharma
Open to work1 years experienceData Scientist
Data Scientist | LLM Training & Evaluation | AI/ML Engineer
Standing Rank
Rank Not Available
Developer Badges
No badges earned yet
Skills & Technologies
28 skillsWork Experience
Data Scientist – LLM Training & Evaluation
Turing
- Authored 500+ SFT prompt-response pairs for Google Gemini across code generation, mathematical reasoning, and multimodal tasks, directly contributing to frontier model training.
- Constructed RLHF preference datasets by evaluating and ranking model outputs across 3 domains (code, reasoning, multimodal), directly informing Gemini's reward model training pipeline.
- Engineered schema-driven validation systems for structured LLM outputs, cutting data errors by ~30% and improving consistency of AI training data workflows at scale.
- Streamlined annotation workflows by building 2 preprocessing pipelines, saving ~8 hrs/week of manual review and ensuring data quality across production SFT and RLHF workflows.
IoT Machine Learning Engineer – Intern
Avkalan Labs
- Developed embedded ML inference pipelines on 2 ESP32 edge devices, achieving real-time data processing with <100ms end-to-end latency.
- Reduced memory footprint by ~40% and cut execution latency on ESP32 devices by profiling and optimizing C++ runtime for constrained ML inference.
- Integrated ESP32 edge devices with a cloud backend over REST, enabling continuous IoT telemetry ingestion across 2 concurrent device streams.
Projects
Taskchain Agent Orchestrator
- Architected a multi-agent workspace with a central orchestrator routing between a Repository Onboarding Agent and a ReAct Coding Agent, cutting onboarding time by ~60%.
- Engineered a sandboxed ReAct Coding Agent with Docker-isolated execution and prompt-injected steering constraints, achieving 100% code isolation for safe, controlled generation.
- Implemented hybrid RAG over ChromaDB + SQLite FTS5, indexing 10,000+ repo artifacts (code, docs, GitHub issues) for contextual agent reasoning.
- Abstracted LLM integrations into a provider-agnostic layer across 4 providers (OpenAI, DeepSeek, Ollama, Nebius) with a Monaco-based UI featuring streaming logs and real-time agent state.
- Deployed on Google Cloud Run; reduced context retrieval latency by ~35% and achieved <2s average agent response time across 3+ LLM providers.
IoT Sensor Data Platform
- Developed FastAPI REST APIs handling real-time data ingestion from up to 4 concurrent IoT sensor devices with <200ms response times and 99%+ uptime.
- Architected a 4-service microservices backend for multi-source sensor pipelines, with decoupled components and real-time telemetry ingestion.
- Modeled a PostgreSQL schema for time-series sensor storage, enabling efficient querying and historical data retrieval across 4+ device streams.
Leaderboard Standings
Leaderboard Position Pending
Global test scores, peer standing percentiles, and algorithm leaderboard ranks are updated dynamically.
Assessment Highlights
Assessments Not Completed
Coding evaluations, system assessment results, and conceptual score badges will appear here after taking a test.
AI Collaboration Score
AI Collaboration Score Pending
Developer coding behavior, assistant cooperation, and AI pair-programming indicators are evaluated during live coding sessions.
Role Compatibility Profile
Role Compatibility Analysis Pending
Custom matching reports, candidate role compatibility percentiles, and core engineer strength profiles are processed once conceptual code screenings are complete.
Achievements
1st Prize, Ware Wizard Hackathon
Won 1st Prize at Cognizance IIT Roorkee's Ware Wizard hackathon.
About Details
Professional Bio
Data Scientist specializing in LLM training, evaluation, and agentic workflows. Experienced in building scalable AI pipelines, RAG systems, and IoT machine learning solutions. Passionate about developing efficient, high-performance AI applications.
B.Tech in Civil Engineering
Indian Institute of Technology (IIT) Roorkee (2021 - 2025)