Open to Backend & AI roles

Dilip Kumar · Backend & AI Engineer · Hyderabad, India

I build reliable backends and AI that doesn't make things up.

Two years shipping production systems with Python, FastAPI, Go and Node.js. Currently a Full-Stack Engineer at Stryv.ai.

See my work Interview my resume

Timeline · 2003 → now

    2003 born → 2020 Class 10 → 2022 Class 12 → 2022 BCA at Kurukshetra University → 2024 first job, NexGen Elit → 2025 graduated → 2025 Full-Stack Engineer at Stryv.ai → next: open to backend & AI roles.

    Scroll

    01About

    A bit about me

    Illustrated portrait of Dilip Kumar

    I studied computer applications at Kurukshetra University in Haryana and joined my first engineering team in June 2024, before I graduated. Since then I've spent two years at early-stage startups, owning auth, payments and AI features end to end.

    I care about systems that behave predictably under load, APIs that are boring to integrate with, and AI features that can show their evidence. Next, I want a team with larger scale and stronger engineering practices. Meanwhile I practice DSA on NeetCode and LeetCode and work through system design.

    160K

    documents a month through an AWS verification pipeline I contributed to

    ~90%

    verification accuracy on that pipeline

    100K+

    users on a CRM I helped build with a 20+ person team

    2 yrs

    shipping production backends, since June 2024

    Education

    BCA, Kurukshetra University

    Govt. College Chhachhrauli · 2022 — 2025

    Recognition

    2nd place, state-level exhibition (150+ colleges)

    College Roll of Honour

    Certifications

    Web Software Development, Altos FITECH 101 University

    Node.js, Simplilearn

    02Experience

    Where I've shipped

    01 / 02

    Jul 2025 — Present

    Hyderabad

    • Python
    • FastAPI
    • PostgreSQL
    • pgvector
    • AWS
    • JWT

    Full-Stack Engineer @ Stryv.ai

    Backend and AI features for a vendor-compliance platform.

    • Building the AI automation module: hybrid vector + keyword retrieval on PostgreSQL/pgvector, cross-encoder reranking, verbatim-quote validation, and async pipelines with human-in-the-loop approval. See how it works
    • Built JWT authentication with protected routes, plus OTP password recovery: hashed, expiring OTPs with resend limits and protection against repeated attempts.
    • Contributed to an AWS document-verification pipeline that processed 160K documents in a month at ~90% verification accuracy.
    • Own the backend for assessment workflows: response management, category progress tracking and trust-score calculation.
    • Developed a CRM for sales, leads and customer interactions serving 100K+ users, as part of a 20+ person cross-functional team.

    Jun 2024 — Jul 2025

    Remote

    • Node.js
    • Express
    • Razorpay
    • REST APIs
    • Google Meet

    Full-Stack Engineer @ NexGen Elit

    My first engineering team, joined before I graduated.

    • Built a job-portal backend where recruiters post roles and students apply through skill-based assessments, with a dual-sided subscription revenue model.
    • Integrated Razorpay for student test payments and recruiter job-posting subscriptions.
    • Built the Node.js/Express backend for a virtual fitness platform: course browsing, an interactive dashboard and integrated Google Meet sessions.

    03Selected work

    Systems you can step through

    01 / 05

    01Work · Stryv.ai

    Grounded retrieval for compliance automation

    An AI module that only accepts an answer it can back with a verbatim quote from the source document.

    • Python
    • FastAPI
    • pgvector
    • Cross-encoder
    • Async pipelines

      Question → keyword search + vector search (pgvector) → cross-encoder rerank → LLM draft with quote → verbatim-quote check → human approval.

      • Hybrid retrieval. Vectors catch paraphrases. Keyword search catches the exact clause names and numbers that embeddings blur.
      • Rerank before generating. A cross-encoder scores every (question, passage) pair, so the strongest evidence reaches the model first.
      • Verify, then involve a human. A quote that isn't in the source verbatim fails validation. Validated answers still wait for human approval.

      The AI automation module of a vendor-compliance platform: retrieval, reranking, quote validation and an async approval pipeline, in FastAPI on PostgreSQL with pgvector.

      02Side project · Live

      Jugaad Link

      A URL shortener with Redis-cached redirects, password-protected and expiring links, and per-link analytics.

      Live Backend Frontend
      • Go
      • Gin
      • PostgreSQL
      • Redis
      • Docker
      • Next.js

        Browser → Gin API → Redis (hit: 302 redirect) or PostgreSQL (miss: rule checks, 302, then cache).

        • Cache only plain links. Password, expiry and click-limit rules must run on every visit. Caching those redirects would silently bypass them.
        • Record clicks asynchronously. Analytics never sits on the redirect's critical path.
        • Fail to a friendly page. Disabled, expired and over-limit links redirect to an explanation instead of a bare error.

        A Go (Gin, GORM) API on PostgreSQL and Redis with JWT auth, custom slugs, QR codes, guest sessions, a Chrome extension, and click analytics by device, browser, country and referrer. Next.js frontend.

        03Side project

        Atlas

        A multi-tenant Slack and GitHub assistant: OAuth 2.0, per-workspace webhooks, and a Gemini agent that calls tools.

        Code
        • Python
        • FastAPI
        • PostgreSQL
        • Gemini
        • OAuth 2.0
        • Next.js

          Slack event → FastAPI → tenant lookup and token decryption in PostgreSQL → Gemini function call → GitHub tool → reply in thread.

          • Encrypt provider tokens at rest. Slack and GitHub tokens are Fernet-encrypted in PostgreSQL and decrypted only for the tenant that sent the event.
          • Rotate refresh tokens. They're stored hashed, rotated on every refresh and kept in HTTP-only cookies.
          • Modular monolith first. One deployable with clear module boundaries (auth, integrations, orchestrator), to split only when scale asks for it.

          A FastAPI modular monolith with a Next.js 15 dashboard: OAuth 2.0 login (Google, GitHub, Slack), organization and workspace tenancy, per-tenant webhook ingestion, and a Gemini orchestrator with function calling and a tool dispatcher. PostgreSQL via async SQLAlchemy and Alembic.

          04Side project · Live

          ContriHub

          Helps developers find beginner-friendly open-source projects, with AI repo explanations and learning roadmaps.

          Live Code
          • Go
          • Gin
          • PostgreSQL
          • Redis
          • Groq
          • Next.js

            Next.js → Go API → Redis rate limit → Redis cache or GitHub API → Groq LLM → 200, or 429 when the limit is hit.

            • Rate-limit every AI endpoint. Redis limits by IP and by token, so one client can't burn the LLM budget or the GitHub quota.
            • Cache the GitHub proxy. Repeat lookups are served from Redis instead of spending GitHub API calls.

            A Go (Gin, GORM) backend on PostgreSQL and Redis: GitHub OAuth, a caching GitHub API proxy, Groq-powered repo explanations, roadmaps and project matching, and persistent AI chat history. Next.js frontend with filters for language, labels like good-first-issue, topics and stars.

            05More

            Also built

            A prototype, daily practice, and everything else.

            Prototype

            Document extraction pipeline

            Serverless OCR + LLM extraction on AWS: Textract for OCR, Bedrock for structured fields. It was never deployed to production.

            • Python
            • Textract
            • Bedrock

            Practice

            DSA in C++

            Regular NeetCode and LeetCode practice, with every submission public.

            Submissions

            Everything else

            On GitHub

            Earlier projects, experiments and the source of this site.

            Dilip-Kashyp

            04Skills

            Every skill, with proof

            Languages

            Backend

            AI & retrieval 1.5y

            Data & cloud

            Backend patterns

            Frontend

            05Ask

            Interview my resume

            A tiny retrieval engine that runs entirely in your browser, over facts I wrote about myself. No LLM and no API key. If it can't find a grounded answer, it says so instead of guessing.

            1. 1

              Tokenize your question and fix typos against the vocabulary

            2. 2

              Score 50 hand-written facts with BM25

            3. 3

              Blend in trigram similarity for fuzzy matches

            4. 4

              Below the confidence threshold? Say “I don't know”

            resume.bot

            answers only from Dilip's resume

            BM25 + trigram

            Hi! Ask me about Dilip's experience, projects, stack or availability. Try a suggestion below or type your own question.

            06Contact

            Have a backend or AI role in mind? Let's talk.

            info@dilipkashyap.in
            • Usually replies within 24 hours
            • Hyderabad, India · IST
            • Open to Bengaluru, remote or relocation
            LinkedIn GitHub

            Prefer JSON? curl it

            dilipkashyap.in/api

            $ curl -s dilipkashyap.in/api/me.json

            200 OK · application/json

            {
  "name": "Dilip Kumar",
  "role": "Backend & AI Engineer",
  "experience_years": 2,
  "current": {
    "title": "Full-Stack Engineer",
    "company": "Stryv.ai",
    "since": "2025-07"
  },
  "location": "Hyderabad, India",
  "open_to": ["Backend Engineer", "AI Engineer", "Full-Stack Engineer"],
  "work_locations": ["Hyderabad", "Bengaluru", "Remote"],
  "focus": ["APIs & auth", "Caching & rate limiting", "RAG pipelines"]
}

            Real static JSON served by this site. Try it in your terminal.

            Esc
            ↑↓ move ↵ select Unknown text goes to the resume bot