Parth Nuwal
Summary
AI Engineer focused on LLM applications, agent systems and the backend infrastructure that makes them reliable. Built X-101, a conversational analytics platform that converts natural language into deterministic dashboards, statistical models and ReAct-driven insights. Comfortable owning systems end-to-end– from prompt design to schema modeling to production deployment.
Experience
Software Engineer InternMar 2026 — Present
AtTheRate.ai, Singapore (Remote)
- Parallelized large-scale campaign reporting pipelines over BullMQ-based async processing, reducing query latency by 30% on high-volume datasets.
- Built backend logic for campaign recommendation systems, including bid and budget adjustments at the placement level (Flipkart platform integration).
- Contributed to system design for Flipkart onboarding (50K+ platform users), including white-labelling architecture and access control across agency vs client permissions.
Business Intelligence InternAug 2024 — Apr 2025
Aspire FinTech Technologies, Singapore (Remote) · LOR
- Implemented 400+ dbt tests (freshness, completeness, consistency, nullability, logic checks) across 30+ core tables, raising data quality score from 30 → 60+ and making downstream analytics demonstrably trustworthy.
- Performed SQL-based data model updates and schema changes, collaborating with stakeholders to align transformations with how the business actually used the data.
- Developed Tableau dashboards on top of the cleaned data layer for campaign performance tracking and business reporting used by growth teams.
Projects
Python, FastAPI, DuckDB, NetworkX, scikit-learn, statsmodels, LiteLLM, Next.js, GCP
- Built an end-to-end system where users upload CSVs and ask plain-English questions, returning dashboards and statistical model outputs backed by parameterized SQL on DuckDB — all execution, joins, and aggregation handled by deterministic Python, never the LLM.
- Architected around the principle “LLMs do language; code does data” — the LLM plans the question, but every SQL query, join, and statistical computation runs through a typed Python execution layer with full schema validation.
- Built a NetworkX-backed semantic graph mapping business concepts to schema columns, with BFS path-finding to inject verified column mappings into the planner — eliminating hallucination on schema references.
- Achieved replay-deterministic execution via JSONL pipeline logging and a deterministic validator (PASS / WARN / FAIL); validated through a 400+ case modular test suite across pipeline stages.
FastAPI, Redis, MongoDB, PyABSA, Hugging Face
- Built a modular NLP pipeline using aspect-based sentiment analysis to extract sentiment, aspects, and intent from review datasets, producing structured outputs for downstream dashboards.
- Deployed scalable backend with Redis-based rate limiting and a monitoring dashboard for tracking system usage and performance.
Skills
- Data & Storage
- SQL, DuckDB, dbt, MongoDB, Pinecone, Chroma, NetworkX
- Backend & Pipelines
- Python, FastAPI, REST APIs, Async Processing, BullMQ, Redis
- ML & Data Science
- PyTorch, scikit-learn, statsmodels, scipy, pandas, NumPy
- Infra & Cloud
- GCP, Docker, Git, n8n
- LLM & AI Systems
- LLaMA 3, Gemini, RAG, Multi-Agent Systems, ReAct, LLM Evals, Langfuse
- AI Frameworks
- LangChain, Pydantic, Hugging Face, Google ADK
Education
B.Tech in Computer Science (Data Science)Jul 2022 — Apr 2026
Swami Keshvanand Institute of Technology (SKIT), Jaipur · CGPA: 9.45 / 10.00
Signals
- Finalist — Smart India Hackathon (SIH) 2025
- Participant — Amazon ML Summer School 2024