Manideep Munjampally
Manideep Munjampally
Open to internshipsClass of 2028 · Hyderabad, India

ManideepMunjampally

AI Systems Engineer & Full-Stack Developer

I’m a computer science undergrad at CBIT who ships production software, and I’m going deeper into ML and agentic AI.

See the workResume
HackerRank Orchestrate
#93/ 3,062
Top 3.1% · September 2026
CGPA
9.74/ 10
Computer Science, CBIT
In production
2apps
Deployed and public on Vercel
Algorithms
200+
Problems solved in C++

Education

Computer science at CBIT.

Chaitanya Bharathi Institute of Technology

Computer Science and Engineering · Hyderabad

Class of 2028
CGPA
9.74/ 10
Core coursework
Data structuresAlgorithmsDBMSObject-oriented design

Selected work

Five systems, each with its architecture on show.

A competition-ranked AI agent, two hackathon team builds, and two solo apps running in production. The panels are live: switch stages and routing modes to see how each pipeline behaves.

01Solo buildFinancial affordability AI · HackerRank Orchestrate 2026

Buy or Wait?

Source

A hybrid AI financial agent that decides whether a user can safely afford a purchase.

Hybrid AI architecture

A strict separation between language understanding and financial reasoning: LLMs extract facts from unstructured evidence, while deterministic systems handle reconciliation, cash-flow simulation, safety validation, and final decision ranking.

Evidence reconciliation

A deterministic reconciliation layer resolves conflicting financial evidence across transactions, messages, and images. LLM output is treated as untrusted evidence and reconciled under explicit rules for amendments, cancellations, newer evidence, and settled transactions.

90-day simulation & decision planning

A cash-flow simulator projects recurring income and expenses while holding the user's minimum required balance, computing the maximum safe payment and the earliest safe purchase date. Candidate strategies — full payment, installments, waiting, spending adjustments — are generated, validated, and ranked deterministically.

  • Python
  • Gemini 3.6 Flash
  • Groq
  • Pandas
  • LLM extraction
  • Deterministic simulation
Affordability decision engineInteractive

The LLM interprets the financial evidence. Deterministic systems establish financial reality and make the decision.

Inspect pipeline stage
4-tier evidence reconciliationDeterministic

A deterministic resolution layer applies explicit precedence rules across amendments, cancellations, newer evidence, and settled transactions to construct canonical ground truth.

Conceptual pipeline · strict separation
  1. Unstructured evidence
  2. LLM fact extraction
  3. Evidence reconciliation
  4. Canonical financial state
  5. 90-day simulation
  6. Safety validation
  7. Payment plan
  8. Final decision
Development results
Status accuracy
68 → 88%
Payment method
72 → 92%
Earliest date
21 / 25

Measured on the project’s evaluation sample during development.

02Team project · voice pipeline and pitch leadAgricultural advisory AI

KrishiCFO

Source

A 3-agent adversarial AI debate platform giving Indian farmers data-driven crop price advice.

Team project & architecture

Built with team Aurexis-Error-404: a commodity price advisory platform powered by a 3-agent adversarial debate system.

Adversarial market debate

Llama 3.1 generates real-time market analysis debates between bullish, bearish, and neutral agents, so the advice a farmer receives has been argued against before it is given.

Voice pipeline & demo leadership

Led the real-time voice pipeline — Whisper STT for regional dialect audio input, ElevenLabs TTS for natural speech output — and led the final pitch and live demo.

  • Python
  • FastAPI
  • React
  • Llama 3.1 via Groq
  • Whisper STT
  • ElevenLabs TTS
3-agent debate engineLlama 3.1 via Groq
Inspect market agent
Bullish · price surge

Local mandi arrivals are down 18% with festival procurement surging. Suggests holding inventory 2–3 weeks to capture anticipated 12–15% upside.

Multilingual voice pipeline
Whisper STTRegional dialect audio in
Groq LPULlama 3.1 adversarial debate
ElevenLabs TTSNatural speech out
03Team of 5 · backend implementorTiered AI inference system

JARVIS

Source

Re-architected from a single cloud endpoint to a tiered Gemini + Groq pipeline, with local Qwen via Ollama for offline resilience.

Team project & role

Built with a 5-member team for a hackathon submission, with Manideep as backend implementor — responsible for multi-tier LLM orchestration, inference routing, and local model fallback.

The engineering challenge

The initial setup could not meet real-world latency and cost requirements. A single cloud model meant frequent rate delays, high per-query cost, and total dependency on connectivity.

Architectural resolution

Re-engineered the backend into a tiered pipeline: standard queries through Gemini for high context, high-speed bursts offloaded to Groq, and Qwen embedded via Ollama as a deterministic local fallback when the cloud is unreachable.

  • Python
  • Gemini
  • Groq
  • Ollama
  • Qwen
Inference fallback simulatorInteractive
Test runtime routing
Tier 1 · GeminiCloud reasoning, high context310ms
Tier 2 · Groq LPULow-latency fallback under loadStandby
Tier 3 · Ollama + QwenLocal node, no network dependencyOffline ready
Telemetry
Endpoint monitor online. Cloud endpoints active.
04Solo buildFull-stack web app

HabitTracker

Open live app

Real-time state synchronization and streak analytics on Supabase, with an activity-driven recommendation engine.

Application architecture

A full-stack habit tracker in React and TypeScript, deployed on Vercel, using Supabase for relational persistence and real-time state replication across client sessions.

State synchronization & streaks

TanStack Query handles optimistic mutations and cache invalidation for zero-flicker streak updates. Completion consistency, historical trends, and daily momentum are visualized with Recharts.

Recommendation engine

Analyzes longitudinal activity history to identify drop-off patterns and surface tailored schedule changes, reinforcing streaks before fatigue sets in.

  • React
  • TypeScript
  • Supabase
  • TanStack Query
  • Recharts
  • Vercel
System capabilitiesProduction live
Real-time sync
PostgreSQL row-level security with Supabase Realtime pub/sub.
Streak engine
Deterministic streak tracking with SVG analytical charts.
Adaptive recommendations
Behaviour analysis suggesting habit pacing adjustments.
05Solo buildProductivity tool

Attendance Analyzer

Open live app

Automates attendance shortage analysis and calculates exactly how many classes are needed to clear the threshold.

The problem

Students face attendance quotas with real penalties for dropping below the minimum. Doing the arithmetic by hand across several courses is error-prone and says nothing about where the trajectory ends up.

Mathematical modeling

A deterministic model takes attended lectures, total conducted sessions, and the target percentage, and returns the exact number of consecutive classes required to exit shortage.

Implementation & deployment

React, TypeScript, and Tailwind CSS for instant reactivity with no calculation round-trip. Deployed to production on Vercel.

  • React
  • TypeScript
  • Tailwind CSS
  • Vercel
Calculation logicZero round-trip
Target equation
classes_needed = ceil((T × total − attended) / (1 − T))

Where T is the required threshold — 0.75 or 0.85 in most institutions.

Margin of safety

Computes how many classes can still be missed while keeping attendance strictly above the institutional cutoff.

Recognition

Tested in competition.

HackerRank Orchestrate

September 2026 · solo entry with Buy or Wait?

Repository

Final placing

#93/ 3,062

Top 3.1%Score 67.7 / 100
Registered
35,712
Shipped a build
3,721
Took the AI interview
3,062

  • 24–36 hour builds

    Hackathons

    KrishiCFO: led the multilingual voice pipeline, then the final pitch and live demo. JARVIS: backend implementor for tiered LLM routing with an offline fallback.

    See both builds

  • LeetCode · C++

    200+ problems solved

    Most of the depth is in dynamic programming, game theory, and divide and conquer.

    Profile

Certifications

Formal grounding in AI.

Coursework behind the pipelines above: one programme completed, one in progress.

  • NPTEL

    IIT-led national programme

    Completed

    Programming with Generative AI

    Prompt engineering, foundational LLM architectures, API integration, and generative programming workflows.

    • Prompt engineering
    • LLM architectures
    • API integration

  • Apna College

    Online programme

    In progress

    AI & Machine Learning

    Classical ML algorithms, deep learning fundamentals, statistical evaluation, and computer vision and NLP concepts.

    • Classical ML
    • Deep learning
    • Model evaluation
    • CV & NLP

Stack

Tools I reach for.

Agentic IDEs are where I build: Claude Code and Antigravity. Gemini, Groq and local Ollama models run inside the pipelines I have shipped.

AI dev tools
  • Claude Code
  • Antigravity
  • ChatGPT
Models & inference
  • Gemini
  • Groq
  • Ollama
  • Whisper
  • ElevenLabs
Languages
  • Python
  • TypeScript
  • C++
Build & ship
  • Next.js
  • React
  • FastAPI
  • Supabase
  • Vercel
  • Git

Content creation

Strategic Timeout.

A cricket channel I create, host, and edit myself in Premiere Pro and After Effects.