Software Engineer @ Telecrm · Gurugram, IN

Lakshya Kamal
Prakash

>

I build AI systems and backend infrastructure that hold up in production — LLM agents, RAG pipelines and REST APIs serving 70,000+ users and thousands of daily active callers. Published at IEEE OTCON 2025, silver medalist in B.Tech (AI) at AMU, and one of the first IndiaAI Fellows.

View work Get in touch
  • 70,000+ users on assistants I shipped
  • 82.5% support tickets auto-resolved
  • 94.97% face-recognition accuracy (IEEE)
  • 9.103 CPI — B.Tech AI, silver medal
01 — Selected work

Things I've built

Full-stack systems, from auth internals to deep-learning pipelines.

Featured

URL Shortener & Link Analytics Platform

Live Repo

A production-shaped link platform: 28 REST endpoints across 6 modules on a layered controller → service → repository architecture, paired with a 16-component Angular 20 SPA built on standalone components and signals.

  • Auth internals. JWT with short-lived access tokens and rotating refresh tokens (SHA-256 hashed, httpOnly cookies); single-flight refresh stops stampedes when concurrent requests hit a 401 together.
  • Privacy-preserving analytics. Salted SHA-256 IP hashes enriched with user-agent + GeoIP for device / browser / OS / country breakdowns — no raw IPs at rest.
  • Hardened redirect path. Protocol allowlisting (blocks javascript: / data: XSS), private-IP filtering for SSRF, a reserved-alias list, and IP + account-keyed rate limiting.
  • TypeScript
  • Express 5
  • Prisma
  • PostgreSQL
  • Angular 20
Research · IEEE OTCON 2025

Facial Recognition for Secure Attendance, with Anti-Spoofing

Paper

An end-to-end recognition system — Deep CNN, HOG, FaceNet embeddings and an SVM classifier — hitting 94.97% accuracy on a real-time dataset of 2,500 students.

  • FASNet-based anti-spoofing detects photo and video replay attacks via per-frame probability scoring with binary cross-entropy loss.
  • Trained and validated against the LFW benchmark; real-time DB sync generates attendance spreadsheets and syncs across devices.
  • Python
  • CNN
  • FaceNet
  • OpenCV
  • scikit-learn
Full-stack

Wanderlust — Travel Accommodation Marketplace

Repo

An Airbnb-style marketplace on a classic MVC split — 14 route handlers and 11 server-rendered EJS views sharing reusable layouts and partials.

  • Session auth via Passport.js persisted in MongoDB (connect-mongo), so logins survive restarts — with post-login redirect restoration.
  • Resource-level authorization middleware (isLoggedIn, isOwner, isReviewAuthor) enforced server-side on every mutating route.
  • Centralized Joi validation, a custom ExpressError class, and an async wrapper funnelling every rejected promise into one error middleware.
  • Node.js
  • Express
  • MongoDB
  • Mongoose
  • EJS
Earlier work college-era projects — kept for the record

  • Tic-Tac-Toe with A.I.

    Minimax-style opponent that makes a human win genuinely hard.

  • Slot Machine — JACKPOT

    The casino jackpot machine, rebuilt in Python.

  • Hangman

    Word-guessing game with a seven-life budget and per-letter scoring.

  • Computer Automation

    Speech-driven virtual assistant that executes local machine tasks.

    private
02 — Experience

Where I've shipped

  1. Jan 2025 — Present Gurugram, Haryana, IN

    Software Engineer

    Telecrm (Flamon Cloudtech Pvt. Ltd.)

    • MCP server. Built on Telecrm's app server, exposing enterprise-creation workflows as callable tools so AI coding agents can provision conversationally.
    • AI call & lead summarization. LLM system that cut insight generation from a week to 30 minutes across 1,095 enterprises, with 24/7 caller-recording analysis.
    • Customer success chatbot. LangGraph + RAG, autonomously resolving 82.5% of post-onboarding tickets.
    • Telecrm Intelligent Assistant. Built on the OpenAI Agents SDK, letting 70,000+ users query CRM data in natural language.
    • Backend APIs. Co-owned endpoints serving 7,000+ DAU, peaking around 10–20 RPS on core routes.
    • Custom Actions. Shipped end-to-end in Angular, TypeScript and Node.js — HLD/LLD and scalability were mine.
    • LangGraph
    • OpenAI Agents SDK
    • RAG
    • Node.js
    • Angular
    • TypeScript
    • MCP
  2. Feb 2024 — Jun 2024 Delhi, IN

    Web Development Teaching Assistant

    Stealth Startup

    • Taught JavaScript, HTML/CSS, MySQL, MongoDB, Express, Node and React — rated 4.32/5 by students.
    • Folded real-world CRUD examples into the curriculum, lifting project completion rates by 18%.
    • JavaScript
    • React
    • Node
    • Express
    • MongoDB
    • MySQL
03 — Toolkit

What I work with

Languages

  • Java
  • TypeScript
  • JavaScript
  • Python

Frameworks & Libraries

  • Angular
  • Node.js / Express
  • NestJS
  • Spring Boot
  • PyTorch
  • TensorFlow
  • Pandas
  • NumPy
  • Matplotlib

Databases & Caching

  • PostgreSQL
  • MongoDB
  • MySQL
  • Redis

Cloud & DevOps

  • AWS
  • Docker

AI & Machine Learning

  • Machine Learning
  • Deep Learning
  • Generative AI
  • Data Science
  • RAG
  • LLM Agents

Core CS

  • Data Structures & Algorithms
  • DBMS
  • Low-Level Design
  • High-Level Design
04 — About

Absorb, then radiate

I'm a software engineer working where applied AI meets real backend systems. Day to day that means agentic tooling, retrieval pipelines and the unglamorous API work that keeps them honest under load.

I finished a B.Tech in Artificial Intelligence at Z.H.C.E.T, Aligarh Muslim University with a CPI of 9.103 — graduating as a silver medalist and one of the first IndiaAI Fellows. My work on anti-spoofing facial recognition was published at IEEE OTCON 2025.

Currently in Gurugram, building at Telecrm. Open to conversations about backend, AI engineering and anything that has to survive production traffic.

Full résumé LeetCode
05 — Contact

Let's build something

Roles, collaborations, or a question about any of the above — all welcome.

Emaillakshya.kp.25@gmail.com WhatsApp+91 73024 12557 LinkedIn/in/lakshya-kamal-prakash GitHub@Lakshya-KP