Backend & Systems Engineer · Agentic AI · Quantitative & High-Performance Computing
B.Tech. Computer Science & Applied Mathematics @ IIIT Delhi (2023–2027)

I build fast, reliable systems — from OS internals and distributed pipelines to LLM agents and parallel numerical solvers.

Systems-oriented engineer with a strong CS + applied-math foundation (DSA, OS, DBMS, Computer Networks, Parallel Programming, Probability & Stochastic Processes, Convex Optimization). I like problems where correctness, latency, and scale all matter at once — distributed backends, concurrency, and numerical computing — and I'm currently deep in agentic AI and quantitative / high-performance systems.

• 🛠️ Backend & distributed systems — FastAPI, Spring Boot, Kafka, REST microservices over TCP/IP.
• ⚡ Systems & HPC — C/C++ on Linux, multithreading, OpenMP/MPI, OS internals.
• 🤖 Agentic AI — LLM orchestration with LangGraph/LangChain, RAG over vector embeddings, and MCP — with deterministic guardrails where money moves.
• 🧠 Applied ML — deep-learning computer vision (YOLOv8, OpenCV), model evaluation with labelled eval sets, NumPy / Pandas / scikit-learn.
• 📈 Quantitative computing — parallel PDE solvers and performance-critical numerical modeling.
• 🧩 Codeforces Specialist (max 1458) — 750+ problems solved (550+ LeetCode, 200+ Codeforces).
• 🏆 IMC Prosperity 4 Finalist — World Rank 823 · All-India Rank 175 · Algorithmic Rank 278.

💼 Open to SDE · Backend · Quant · AI — internships & full-time (Class of 2027).

• Shipped a production gamified campus-ambassador platform for 100+ users — dual portals, live leaderboards, and 20+ REST APIs in TypeScript / Next.js on a 17-table Supabase (PostgreSQL) schema across Agile sprint cycles.
• Eliminated concurrency defects under peak load via atomic transactions and row-level locking; enforced secure coding (input validation, parameterized queries) and validated releases with 433 automated tests across 21 suites.

• Accelerated the explicit PDE solver ~10× on multicore CPUs with OpenMP (persistent teams, cache-friendly access) and scaled it across nodes with MPI (domain decomposition, ghost-cell exchange), benchmarked against a serial baseline.
• Implemented Explicit, Implicit, and Crank-Nicolson finite-difference solvers for the Black-Scholes option-pricing PDE (C++/Python), benchmarked against analytic prices; exploiting Crank-Nicolson's unconditional stability cut average solve time from 2.69 s to ~6 ms at equal-or-better accuracy.

• Engineered asynchronous inter-service communication between C++ and Java (Spring Boot) microservices using Kafka over TCP/IP in a distributed, mission-critical defense system, improving processing efficiency by 60%.
• Trained a YOLOv8 deep-learning object-detection model for maritime vessels on HRSC2016 satellite imagery to 0.83 mAP@0.5, with OpenCV-based image preprocessing and a custom XML→YOLO annotation parser.
• Streamlined C++ Linux service deployments into a CI/CD pipeline, eliminating manual integration errors across test environments.

• Multi-tenant agentic AI platform automating the dominant support workload of food-delivery and e-commerce brands — order disputes and refunds: an autonomous LLM agent on an 11-node LangGraph state graph adjudicates every ticket against the brand's uploaded policy via per-tenant RAG (pgvector) with citations — across 68 REST APIs, a Kafka pipeline (idempotent consumer, dead-letter queue, transactional outbox), and 4 role-gated portals behind RBAC.
• Fraud-resistant money path: deterministic guardrails evaluated outside the model (prompt injection cannot move money), server-side evidence verification, human-in-the-loop escalation, and idempotent execution with per-session locks — retries and redeliveries never double-pay.
• Red-teamed twice (5-agent adversarial pass + 20-case hostile E2E sweep) — 11 exploits found, fixed, and pinned as regressions; 290+ tests and a 44-case labelled offline eval gate every push in CI: 88.6% auto-resolution, zero guardrail violations — deployed live on Render.

• Team project (4 devs): full-stack ERP + LMS serving 250+ students and faculty — 40+ REST APIs, analytics dashboards, and event-driven email workflows over a 19-table PostgreSQL architecture, including a Wi-Fi-based real-time attendance system (MAC/BSSID detection).
• My ownership: the gamified feedback & points system end to end — leaderboard (API + UI), admin feedback analytics dashboard with privacy controls, student feedback UX, and automated cron + email pipelines (verify: git log --author="Akshat").

• Built finite-difference solvers for the Black-Scholes PDE with MPI and OpenMP, using domain decomposition and ghost-cell communication.
• Benchmarked runtime, stability, and accuracy across serial, MPI, and OpenMP variants with supporting analysis.

• Programmed an MLFQ-based scheduler daemon inside a custom C shell (fork/exec/wait system calls), managing 30+ concurrent processes with adaptive priority scaling (~12% CPU-utilization gain under load).
• Added semaphore-based synchronization and timing-driven deadlock detection, eliminating race conditions under heavy concurrency.

Languages

Backend & Systems

Frontend & Data

AI & ML

DevOps & Tooling

• 🏅 Codeforces Specialist — max rating 1458; 750+ DSA problems solved (550+ LeetCode, 200+ Codeforces).
• 🏆 IMC Prosperity 4 Finalist — IMC Trading's global algorithmic trading competition · World Rank 823 · All-India Rank 175 · Algorithmic Rank 278.
• 🎓 SMP Recognition Award (2025–26) — Student Mentor, IIIT Delhi's Student Mentorship Program.
• 🚀 State-level qualifier, Smart India Hackathon 2024.
• 🔢 Mathematics topper — CBSE 2021 with a perfect 100/100.
• 🎤 Delivered 3+ tech workshops at ESYA (IIIT Delhi's tech fest) to 200+ participants with 90%+ positive feedback.