Kunal
Aggarwal
prev_roles = ["Backend AI Intern @ FlyRank AI", "SWE Intern @ Urlyte"]
freelance = "AI Model Analyst @ Deccan AI Experts"
status = "open to roles"
focus = "agentic systems, LLM evaluation, prod backend"
leadership = "McKinsey Forward Alumni · 2026"
obsession = "the almost-correct outputs. that's where it gets interesting."
➛ FastAPI · LangGraph · Docker · AWS EC2 · PostgreSQL
➛ hitting walls at 2am so production systems don't.
01 / about
Who I Am.
I'm Kunal Aggarwal, a CS undergrad at Vellore Institute of Technology who has spent the last year shipping production AI systems across multiple roles. I interned as a Backend AI Engineer at FlyRank AI, building low-latency FastAPI services and LLM-powered pipelines on AWS EC2. Before that, I was a Software Engineer at Urlyte, optimising a high-throughput URL redirect service down to 63ms p99 latency. Alongside both, I freelance as an AI Model Analyst at Deccan AI Experts, running RLHF cycles and LLM evaluation across Python and SQL codebases.
My focus is agentic AI systems: multi-step LLM pipelines, tool-calling agents, and retrieval-augmented generation. I have real hands-on time in LangChain, LangGraph, smolagents, and Google ADK and I care about systems that hold up in production, not just notebooks that run clean. I tend to go deep, understanding why a model breaks is more interesting to me than patching the output.
Outside AI work, I build full-stack products with FastAPI, Next.js, Docker, and PostgreSQL, and have a background in blockchain systems and quantitative finance. I am also a McKinsey Forward Program alumnus; structured problem-solving is a skill I use in debugging as much as in strategy.
Agentic AI systems, LLM evaluation pipelines, multi-agent orchestration, RAG. Currently going deep on LangGraph state machines and transformer internals. Building toward MCP (Model Context Protocol) integrations.
India, remote-first. Open to on-site opportunities across India or international remote roles in AI/ML and backend engineering.
LLM alignment & safety, RLHF systems, agentic reasoning, blockchain infrastructure, distributed systems, competitive programming.
kunal.aggarwal.office@gmail.com · +91 8929970321
02 / experience
Where I've Worked.
- Engineered 12 low-latency REST API endpoints with FastAPI; architected a modular service layer handling structured request routing and tool-calling across 6 third-party integrations and 7 backend services, achieving p99 response times of ~270ms under production load.
- Designed normalised PostgreSQL schemas and containerised 5 microservices with Docker; automated build and deployment pipelines via GitHub Actions CI/CD, cutting manual pipeline execution by ~30 minutes per release cycle.
- Investigated production edge-case failures in LLM-powered endpoints and converted recurring failure patterns into reusable evaluation cases, improving agent reliability across iterative releases.
- Authored prompt templates and tool-call schemas that standardised inter-service communication, reducing integration bugs across the distributed backend.
- Built and optimised a high-throughput URL redirect service, driving p99 redirect latency down to 63ms, enabling real-time resolution at scale with minimal overhead.
- Developed REST APIs for custom slug generation and management, reducing new link provisioning time to under 30 seconds end-to-end across server infrastructure.
- Contributed to the user account dashboard backend, implementing account management flows and link analytics endpoints to support user-facing features.
- Conducted 30+ RLHF comparative preference annotations evaluating LLM outputs on correctness, factual grounding, and safety; labels used directly in downstream fine-tuning pipelines.
- Achieved a 96% task approval rate across 26 structured debugging submissions; diagnosed model failure modes including hallucinations and reasoning gaps in Python and SQL outputs.
- Drove a 15%+ reduction in code error rates via systematic edge-case and exploratory analysis of model outputs.
- Mentored 15 junior peers on evaluation methodology and annotation standards, accelerating onboarding and maintaining team-wide quality consistency.
03 / projects
Things I've Built.
Full-Stack AI Product · Fine-Tuned Domain LLM
Production-grade medical Q&A system built end-to-end. Fine-tuned Phi-3 Mini (3.8B) using QLoRA on 100k+ real doctor-patient conversations. Two-pass RAG pipeline: PubMedBERT bi-encoder retrieves from ChromaDB, ms-marco cross-encoder reranks for accuracy. BERTScore F1 = 0.80. FastAPI backend with JWT auth, rate limiting, PostgreSQL + pgvector. Next.js frontend on Netlify. Docker, CI/CD via GitHub Actions, DVC data versioning, MLflow experiment tracking.
Multi-Agent Recruitment System
Production multi-agent recruitment pipeline. A SequentialAgent (Google ADK + LangGraph state machine) orchestrates 4 specialised LlmAgents: JD Parser, Candidate Matcher, Outreach Composer, and Interview Scheduler. Deployed on AWS EC2 via Terraform. Includes a Bias Guardrail agent that flags discriminatory signals in evaluation outputs before candidate ranking.
↗ github.com/koi-bito/HireFlowBlockchain Credential Verification
Fault-tolerant distributed ledger for tamper-proof credential verification. Credentials stored as SHA-256 hashes on Ethereum Sepolia; no central datastore needed. FastAPI + PostgreSQL backend syncs on-chain state via Web3.py event listeners. Decoupled on-chain and off-chain layers for independent scalability and fault isolation. Solidity smart contracts deployed via Hardhat.
↗ github.com/koi-bito/CertChainAI-Powered Profile Generator
LangChain parsing engine that maps natural language speech to structured JSON schemas via embeddings and generative AI, optimised for ATS keyword alignment. Modular async backend integrating OpenAI Whisper for transcription and multi-step LLM pipelines to convert raw audio into formatted, ATS-ready resume documents.
↗ github.com/koi-bito/audio_to_resumeAI-Powered E-commerce Recommender
Full-stack e-commerce platform with a hybrid ML recommendation engine. TF-IDF + KNN with FULLTEXT search and cached REST API endpoints for personalised product recommendations at scale. Secure JWT auth, user behaviour tracking, and Chart.js analytics dashboards.
↗ github.com/koi-bito/koimate04 / skills
My Toolkit.
05 / education
Where I Learned.
06 / achievements
Recognition & Certs.
Completed McKinsey's highly selective leadership and capability-building program. Trained in structured problem-solving, stakeholder communication, data-driven decision-making, and cross-functional collaboration frameworks. Active member of the Forward Alumni Network; a global community of emerging leaders across industries.
featured certifications
also completed
Want the full picture?
Download my resume; updated with all roles, projects, and certifications.
07 / contact
Let's Talk.
I'm actively looking for AI/ML internships and entry-level engineering roles; anything involving LLM systems, agentic pipelines, or backend AI infrastructure. I also explore research directions in LLM alignment and safe agentic systems.
Got an interesting problem? Want to talk RLHF, multi-agent systems, or distributed backends? Or just want to connect; I'm easy to reach.
Message sends directly, no email client needed.