prachi@portfolio — zsh — 90x24

prachi@portfolio:~$ cat about.md

// profile loaded

Prachi Agarwalla . AI/ML Engineer

Building RAG pipelines, agentic AI systems, and applied ML products.

1,857 · LeetCode Knight 3rd · Build4Bharat Hackathon 450+ · DSA Problems Solved

Final-year AI/ML student at UPES Dehradun specializing in computer vision, RAG, and agentic workflows. From building floorplan-digitization pipelines to deploying multi-agent apps at 24-hour hackathons, I focus on turning complex problems into scalable solutions. My approach is guided by a simple motto: "Discipline outlasts motivation." Whether I'm fine-tuning models or fixing edge cases, my goal is always to build systems that work reliably under real-world pressure.

View work → ↓ Résumé Say hello

prachi-agarwalla.vercel.app

stack.json

Technical Skills

Languages

PythonJava TypeScriptJavaScriptSQL

AI & ML

RAG PipelinesAgentic AI Multimodal LLM ReasoningLangGraph Fallback RoutingSHAP Explainability PyTorchOpenCV

AI APIs

Gemini APIGroq API Google GenAI SDK

Frontend

Next.jsReact ViteTailwind CSSFramer Motion

Backend

Node.jsExpress.js REST APIsDrizzle ORMNextAuth

Databases & Cloud

PostgreSQLpgvector HNSWRedisSupabase VercelDockerGit

git log --experience

Work Experience

commit 9f2a1cd

AI Intern Amaippu (Chithiram) ongoing Jul 2026 — Present

Built an end-to-end AI pipeline to convert architectural floorplans from PDF, DXF, and DWG formats into structured vector representations for automated spatial analysis.
Designed WallNet, a DETR/LETR-inspired Transformer architecture with Hungarian matching for direct continuous wall vector prediction.
Improved wall detection performance by 9.8% over the segmentation baseline while eliminating heuristic pixel-to-vector post-processing.

commit 4b7e803

ML Intern (Backend Lead) Sentienta QualityAI Jun 2025 — Jul 2025

Engineered a production OCR-to-LLM extraction pipeline converting unstructured document scans into structured, queryable data
Cleaned real-world datasets and ran EDA & error analysis to improve extraction accuracy
Collaborated with engineers to ship scalable, production-ready ML pipeline solutions

./projects

Projects

Full-Stack · AI Placement Platform

build

SkillSync

AI-driven placement and ranking platform doing semantic candidate discovery. A hybrid ranking engine (70% semantic search + 30% skill-overlap scoring) powers 69+ REST API routes, with 22 PostgreSQL RLS policies enforcing multi-tenant access. A multi-tier AI fallback router (Gemini/Groq) with circuit-breaker monitoring keeps inference reliable.

● 69+ API routes · 22 RLS policies · hybrid ranking

Next.js · PostgreSQL · pgvector · Gemini API · Groq API

code ↗

AI Pipeline · Content Compliance

build

Lumina

A 7-agent LangGraph pipeline automating enterprise content generation from brief to compliance-cleared, localized output, with SSE-streamed real-time progress. A 3-pass autonomous compliance loop (rule-citation + auto-rewrite) raised the catch rate from ~70% to ~95%, validated by 131 tests, with an editorial memory loop learning from human corrections.

● ~95% compliance catch rate · 131 tests · ET AI Hackathon 2026

LangGraph · FastAPI · React · Groq · Gemini

code ↗

AI Pipeline · Multimodal Verification

build

Veridict

Multimodal claim-verification pipeline evaluating evidence sufficiency and claim status for cars, laptops, and packages. Deterministic rule engines enforce manual-review escalation independent of LLM output, keeping the system safe against model drift or hallucination.

● 144 claims processed · 64 evidence-gate overrides

Python · Gemini 2.5 Flash · Google GenAI SDK

code ↗

ML · Candidate Ranking & Explainability

build

Veritalent

Candidate ranking and explainability system built for the India Runs Data & AI Challenge. Combines GBM LambdaRank for ranking, dense retrieval for candidate matching, and SHAP attributions so every ranking decision is explainable rather than a black box.

● GBM LambdaRank · dense retrieval · SHAP

Python · LightGBM · SHAP · Dense Retrieval

code ↗

Agentic AI · Support Triage

build

Triad

Terminal-based multi-domain support triage agent built for the HackerRank Orchestrate hackathon, with a 5-agent architecture routing and resolving tickets across domains. Built and debugged the corpus indexer and retriever powering grounded, hallucination-resistant responses.

● 5-agent architecture · HackerRank Orchestrate

Python · Multi-Agent Orchestration · Retrieval

code ↗

ML · Traffic Demand Prediction

build

Geohash

Traffic demand prediction system built for Gridlock Hackathon 2.0. The key insight was a deterministic (geohash, timestamp) lookup structure paired with per-geohash calibration, feeding a LightGBM model tuned for real-world demand forecasting.

● Geohash lookup structure · per-geohash calibration

Python · LightGBM · Geohash

code ↗

stdout.log

Achievements

1,857

LeetCode Knight — top 6.07% globally, 450+ problems solved across GFG and LeetCode

3rd Place

Build4Bharat, 24-Hour CSI Uttarakhand State Hackathon — Team Lead, built and presented Jan-Saathi (govtech platform) against 147 teams

Top 3,000

Amazon ML Summer School 2026 — selected nationwide

$100

AI4Bharat Hackathon — Round 2 shortlist, AWS credits awarded

Semi-Finalist

ET GenAI Hackathon 2026, The Economic Times Digital — for Lumina

Top 7%

HackerRank Orchestrate, May 2026 — for Triad, a 5-agent support triage system

45-Day Bootcamp

Cisco CWiCS BOOST 6.0 — shortlisted and attended nationwide

./certifications

Certifications

Oracle Cloud Infrastructure 2025 Certified AI Foundations Associate

OracleNov 2025

ID · 323958471OCI25AICFA

Semi-Finalist

ET GenAI Hackathon 2026, The Economic Times DigitalMar 2026

Problem Solving

HackerRankNov 2025

ID · 9B7701A2602C

CCNA: Introduction to Networks

CiscoJun 2026

ID · 63cdc966-19f9-487e-9aef-3e008f46e31a

CCNA: Switching, Routing & Wireless Essentials

CiscoJun 2026

ID · 7b6b74e2-d0ec-46ed-a5dd-b89535be62ec

CCNA: Enterprise Networking, Security & Automation

CiscoJun 2026

ID · ea339d92-cd39-48e9-ba03-3f32a10d5aeb

SQL (Intermediate)

HackerRankNov 2025

ID · BC1AA80F1B75

Data Analytics Essentials

CiscoJun 2026

ID · 31dda895-00bb-4371-923c-34a571c95440

./education

Education

UPES, Dehradun

B.Tech, Computer Science — Artificial Intelligence & Machine Learning

2023 — 2027

8.46

CGPA / 10.00

DAV Public School, MCL JA Dera

Class XII · CBSE

93.1%

DAV Public School, MCL JA Dera

Class X · CBSE

91.6%