§ Coding & Implementation · Reproducible code for thesis methods

Your method, made runnable — implemented by an engineer who has read your draft.

Reproducible implementations for thesis chapters and journal artefacts — Python, R, MATLAB, NS-3, Java, C++. Documented, runnable on a fresh machine, defensible at viva. One named engineer-editor on the file, paired with a PhD subject lead in your area.

01Method-first, not snippet-first. We read your method chapter before we touch a keyboard. The implementation matches your equations, your variables, your notation — not a generic tutorial.
02Reproducible on a fresh machine. Pinned dependencies, seeded randomness, a one-command run script, fixtures and synthetic data. If it does not run on a clean VM, we have not finished.
03Written to be read at viva. Inline docstrings in your supervisor's tongue, a README that walks the examiner from clone → result, and a short walk-through video keyed to your method's three hardest steps.
04Coordination, not chase-up. Five fixed touchpoints. One named engineer. Daily commit log. The chase that broke this service before — gone.

↓ What you receive · Stacks & scope · Coordination · Pricing · Common questions

§ Service note Reopening 2026 · paused since 2021 while we rebuilt the engineering protocol. Read about the coordination changes.

New submission

id · DRAFT-26.04

Choose Service *

No. of Pages *

Document Language *

Document *

Drop file or click to upload

.docx · .doc · .pdf · max 50 MB

Email ID *

First name *

Last name

Phone (with country code) *

Country *

Receive OTP on

SMS Email WhatsApp (coming soon)

I have read the Privacy Policy and agree to the terms.

Note: Due to a delay in downloading reports from Turnitin, some report deliveries may be delayed. We appreciate your patience.

180+

repos delivered

across CSE, ECE, IT, Bioinformatics — 2014–2021

93%

first-pass viva acceptance

examiner ran the code; the code ran

engineer-editors

paired with a PhD subject lead at quotation

5–10d

typical turnaround

for a method-chapter implementation

Code shipped against viva and journal artefacts in Python · R · MATLAB · NS-2 / NS-3 · Java · C++ · PHP · ASP.NET · TensorFlow · PyTorch · scikit-learn · OpenCV · NLTK · Hadoop / Spark · Docker · Git

§ 02 · The repository

What arrives in your inbox — a runnable repo, not a folder of scripts.

A real PhD chapter implementation, delivered Apr 2026 to a Civil Engineering scholar at Anna University. Three artefacts the examiner will touch — README, training loop, and the regenerated table.

~/ppgat main · 47 commits

▸ src/
└─ train.py
└─ models/ppgat.py
└─ data/indore_split.py
▸ data/
└─ raw/ · synthetic + PEMS-BAY
▸ reports/
└─ table-4-2.md
└─ fig-4-3.svg
└─ fig-4-5.svg
▸ docs/
└─ walkthrough.mp4
└─ methods-note.pdf
README.md
Makefile
environment.yml
Dockerfile
.gitignore

A README that walks an examiner from clone to result.

# Pathway-pruned graph attention for traffic flow forecasting

· Implementation accompanying Chapter 4 of A. Iyer, "Spatio-temporal forecasting in semi-urban road networks" (2026).

## Quick start

git clone <repo>
cd ppgat
make env       # creates pinned conda env from environment.yml
make data      # downloads PEMS-BAY + the synthetic Indore split
make train     # 4 GPU-hr on a T4; ~25 min on the M2 fixture
make report    # regenerates Tables 4.2–4.4 and Figs 4.3, 4.5

## Reproducing the thesis figures

· Each table and figure in Chapter 4 has a matching make target. `make table-4-2` runs the held-out evaluation; `make fig-4-5` regenerates the attention-weight heatmap.

## What this implements

A pathway-pruned variant of the GAT layer (Eq. 4.7 in the thesis), with the four pruning rules from § 4.3.2. The training loop, the masking scheme, and the diurnal sampler are all as described in § 4.4 — variable names match the equations, not the original Veličković reference.

## Walk-through video

· `docs/walkthrough.mp4` — 11 min, three hardest steps explained on a whiteboard. Keyed to lines in the codebase.

Variable names match your method. Docstrings cite your equations.

# src/train.py
# Implements the training loop from Chapter 4, § 4.4 (Iyer, 2026).
# Variables follow the thesis notation, not the Veličković et al. (2018) reference.

import torch
from torch import nn, optim
from .models.ppgat import PathwayPrunedGAT     # Eq. 4.7
from .data.indore_split import DiurnalSampler  # § 4.4.2

def train_one(cfg):
    """Train one PPGAT model on one fold.

    Parameters
    ----------
    cfg : Config
        Hyperparameters; see thesis Table 4.1 (page 87).
    """
    g, X, y = load_graph_and_signal(cfg.split)

    # Pruning rules R1–R4 from § 4.3.2.
    # R3 (degree-skew) is the contribution of this thesis;
    # ablation in Table 4.4 reports its effect.
    A_pruned = prune_pathways(g.A, rules=cfg.rules)

    model = PathwayPrunedGAT(
        in_dim=X.shape[-1],
        hid=cfg.hid,
        n_heads=cfg.heads,         # K in Eq. 4.7
        adj=A_pruned,
    )
    opt = optim.Adam(model.parameters(), lr=cfg.lr, weight_decay=cfg.wd)

    sampler = DiurnalSampler(X, y, cfg.window, cfg.horizon)
    for epoch in range(cfg.epochs):
        for xb, yb in sampler:
            yhat = model(xb)
            loss = nn.functional.l1_loss(yhat, yb)   # MAE; § 4.4.3
            opt.zero_grad(); loss.backward(); opt.step()
        log_metrics(epoch, model, sampler.val())

    return model

The exact tables and figures your thesis reports — re-runnable.

## Table 4.2 — held-out MAE on Indore semi-urban split

· `make table-4-2` · seed 42 · single-T4, 4 GPU-hr · last regenerated 2026-04-28 14:02 IST.

Model              | 15-min MAE | 30-min MAE | 60-min MAE
-------------------|-----------:|-----------:|----------:
HA (baseline)      |      3.81  |      4.42  |      5.10
LSTM               |      2.94  |      3.51  |      4.23
DCRNN              |      2.61  |      3.18  |      3.81
Graph-WaveNet      |      2.55  |      3.09  |      3.74
PPGAT (ours)       |      2.41  |      2.92  |      3.55  ←  reported in thesis
PPGAT, no R3       |      2.49  |      3.04  |      3.68     (ablation, Table 4.4)

Lower is better. Bold in the manuscript; reported to two decimals to match journal style.

## Reproducibility

Every number above is regenerated from `make table-4-2`. Random seed is fixed (seed = 42 across torch, numpy, python). Hardware: NVIDIA T4 (16 GB), CUDA 11.8, driver 535. The same script falls back to CPU if no GPU is available — Apple M2 Pro fixture takes ≈ 25 min.

## Bundled with delivery

· Walk-through video · seeded synthetic dataset · environment.yml · Dockerfile · 11-page methods note formatted to your thesis style.

Examiner canclone, run make report, see your thesis tables regenerate live.

Supervisor canread the daily commit log; one click to a side-by-side diff.

You candefend it. Every variable in the code is named the way it is named in your thesis.

§ 03 · Stacks & scope

What we implement, by stack and by what your chapter actually is.

A non-exhaustive list. If your method sits across two of these, we pair the engineer with a second editor — common for ML × bioimaging, statistics × econometrics, networking × ML.

Machine learning · NLP

Thesis methods. Replication studies. Ablations.

PyTorch TensorFlow scikit-learn HuggingFace NLTK spaCy OpenCV

PyTorch, TensorFlow, scikit-learn, HuggingFace, NLTK, spaCy. From a CNN baseline you have outgrown to a paper-fresh transformer variant — in your equations, with your data, against your baselines. Ablation tables built in.

We deliver

Faithful reproduction of your method
Baselines + ablations
Seeded reproducibility
Methods note for thesis

Typical context

CSE / ECE / IT thesis Chapter 4

Statistics · biostatistics · econometrics

R, Python, SPSS — analysis you can defend.

R tidyverse lme4 survival Python pandas statsmodels SPSS Stata

GLMs, mixed-effects, survival, time-series, multivariate. We work in R when your reviewer wants R; in Python when your supervisor wants Python; in SPSS when your university's template demands it. Annotated notebooks, not opaque button-clicks.

We deliver

Cleaned, audited dataset
Annotated analysis notebook
Diagnostic plots
Methods + results draft

Typical context

Public health · economics · psychology

Image, signal & medical processing

OpenCV, MATLAB, ITK — figures examiners can re-render.

MATLAB OpenCV ITK SimpleITK PyDICOM scikit-image PyTorch

Segmentation, classification, registration, frequency-domain analysis. We work in MATLAB when your lab does, in Python when your reviewer asks for it. Figures regenerate from a single command — no PowerPoint freezing of results.

We deliver

Pre-processing pipeline
Trained model + weights
Quantitative tables
Re-renderable figures

Typical context

Bioimaging · radiology · remote sensing

Networking · simulation · MANET

NS-2, NS-3, OMNeT++ — protocol-level studies.

NS-2 NS-3 OMNeT++ MATLAB C++ Tcl

Routing protocols, energy models, AODV/DSR/OLSR variants. We write Tcl, we write C++, we read your CFP. Topology files, mobility models, and a tracegraph script that produces the throughput / PDR / delay plots in your chapter.

We deliver

Protocol implementation
Topology + mobility files
Trace parsers
Throughput / PDR / delay plots

Typical context

CSE wireless · IoT · sensor networks

Big data & backend

Hadoop, Spark, ASP.NET, Java, PHP — runnable systems.

Hadoop Spark Java C++ ASP.NET PHP Docker

When the chapter is a system, not a model. MapReduce jobs, Spark pipelines, ASP.NET / Java backends, dashboards. We containerise; we document the fixtures; we write the seed scripts so the examiner can run a 12-record demo without your full corpus.

We deliver

Working backend / pipeline
Seeded fixtures + demo
Containerised run
Architecture diagram

Typical context

IT · software engineering · informatics

§ 04 · Coordination

One engineer, one subject lead. Five fixed touchpoints. No chase-up.

When this service ran in 2019, the breaking point was always the engineer–scholar handoff — code that did not match the method, results that the supervisor saw for the first time at delivery, examiners who could not run the artefact. The protocol below is what we have rebuilt the service around.

Day 0

Brief

Kick-off call

Forty-five minutes. Engineer + PhD subject lead read your method chapter beforehand. We agree the equations to be implemented, the baselines, the metrics, the data fixtures, and what "done" looks like.

SLA Booked within 24 h of payment · Actors You + engineer + subject lead
Day 2

Plan

Implementation plan

A short document — file tree, equations mapped to functions, baselines list, ablation list, the demo dataset shape, the make-target list, the risks. You sign off before any code is written.

SLA Your feedback within 48 h · Actors Engineer → you
Day 5

Milestone

First runnable

A repo that clones, installs, runs end-to-end on a small fixture and produces a result — not the final number, but a number. You see the architecture. We surface any data or method blockers early.

SLA Same time slot, weekly · Actors You + engineer
Day 8

Results

Numbers + ablations

The real run on your full data. Tables 4.2–4.4 and Figures 4.3, 4.5 regenerated. Your supervisor reads the commit log. We surface any number that disagrees with your manuscript before you read about it from a reviewer.

SLA Your feedback within 3 days · Actors Engineer → you + supervisor
Day 10

Final

Repo handoff + walk-through

Final repo with README, walk-through video, methods note, Dockerfile, and the regenerated thesis tables. One round of revisions included if your supervisor or examiner flags changes.

SLA Revisions within 48 h · Actors Engineer → you

If we miss an SLA

5% credit per day late, off your final invoice.

If you miss feedback

Timeline shifts by the same number of days. We tell you, in writing, on day one.

Engineer change

Only at your request. Otherwise the same person from kick-off to handoff.

Off-channel

Email + WhatsApp + scheduled calls. Daily commit log shared with your supervisor.

§ 05 · The process

Brief to handoff — step by step.

i.

Brief upload

Drop your method chapter, your supervisor's direction, and any data you can share. No data yet? Send the equations and a sample-size description — we will build a synthetic fixture for the first run.
ii.

Quote · within five minutes

We slot the work into a stack and scope band, name the engineer and the subject lead, and email a quotation. No payment until you approve. We tell you who will write the code, not just what it will cost.
iii.

Engineering coordination

Five fixed touchpoints across the timeline (see § 04). Plan by Day 2, first runnable by Day 5, real numbers by Day 8, handoff by Day 10. Adjusted for larger systems (NS-3, big-data backends).
iv.

Repo handoff

A runnable repository, walk-through video, methods note, Dockerfile, and the regenerated tables. One round of revisions included if your supervisor or examiner flags changes.

§ 06 · Pricing

By chapter shape, not by line count.

Code is not paid by the page — and we do not pretend it is. Quotes are scoped at the kick-off call against your method chapter. Floor prices below; the binding number arrives at quotation.

Chapter shape	Scope	From	TAT	What this slab covers
Algorithm replication	1 method · 1 dataset	from ₹18,000	5–7 days	Faithful reproduction of one published method against one of your datasets. Baselines included.	Quote →
Thesis chapter (ML / NLP) Most common	Method + 3 baselines	from ₹32,000	7–10 days	Most common slab. Your equations, three baselines, an ablation, a reproducible repo.	Quote →
Statistics / biostat study	Cleaning + analysis	from ₹22,000	5–8 days	Auditable notebook from raw to results. Diagnostic plots, model checks, methods + results draft.	Quote →
Image / medical processing	Pipeline + figures	from ₹28,000	7–12 days	End-to-end pipeline, trained model, regenerable figures, and a methods note.	Quote →
Networking simulation	Protocol + plots	from ₹26,000	7–10 days	NS-2 / NS-3 / OMNeT++. Topology, mobility, and the throughput / PDR / delay plots in your chapter.	Quote →
System / backend	App + fixtures	from ₹40,000	14–21 days	When the chapter is a working system. Containerised, seeded, demo-ready.	Quote →
Big-data pipeline	Spark / Hadoop	from ₹48,000	14–21 days	Distributed pipelines with seed fixtures and a 12-record demo your examiner can run locally.	Quote →

Repo handoff

Always included. README, Makefile, Dockerfile, env file, walk-through video.

Revisions

One round included after handoff. Subsequent rounds at ₹1,200 / engineer-hour.

Bundled report

Methods note in your thesis style · 6–12 pages · no extra fee.

§ The promise

It runs on a fresh machine. Or we have not finished.

The reason this service paused in 2021 was operational, not technical. Code shipped, but supervisors only saw it at the end; examiners could not always reproduce it; engineers and scholars worked from different mental models of the method. We rebuilt the service around five fixed touchpoints, a paired engineer + PhD subject lead, and a daily commit log. The promise below is what that protocol underwrites.

It runs · A fresh VM clone, make report, your tables regenerate. If not, we fix it on our time.
One engineer · Same person from kick-off to handoff. Paired with a PhD subject lead in your area.
One revision round · After supervisor or examiner feedback, included.
On time · We hit every SLA, or 5% off the invoice for each day late.
Refund · If the repo does not run on a fresh machine within thirty days of handoff for reasons on our side, the work is free.

§ 07 · Voices

Scholars whose thesis code we wrote — before the service paused, and since reopening.

★★★★★

My examiner asked to run the code at viva. He cloned it on his laptop, ran make report, and watched the tables regenerate. He smiled and asked the next question. That's the moment I knew this had been worth the spend.

A. Iyer

Anna University · PhD Civil

Thesis chapter (ML)

★★★★★

I came in with a method chapter and a folder of half-finished notebooks. Two weeks later I had a repo my supervisor could read on a Monday morning and a methods note that ended up as the second half of my Chapter 4. The engineer named the variables the way I had named them in my equations — that detail mattered.

Pranay D.

IIT Madras · PhD ECE

Image processing

★★★★★

Statistics, not ML. I wanted my mixed-effects model done in R because Reviewer 2 had asked, and I'd been doing it in SPSS. Their statistician redid the entire analysis in lme4, sent annotated notebooks, and the diagnostic plots caught a violation I had missed. Paper accepted in revision.

Dr. P. Bhatia

AIIMS · Public health

Statistics

★★★★★

NS-3 work. I had been chasing a freelance for three months. Their engineer sent the protocol implementation in seven days and the throughput plots matched what my supervisor expected from the routing rule. The code is clean enough that my junior is now extending it for his own chapter.

R. Subramaniam

NIT Trichy · M.Tech CSE

Networking

★★★★★

What sold me was the daily commit log shared with my supervisor. He could see the work happening; I never had to explain a delay. The walk-through video saved me twice during my mock viva.

Gurri Bhullar

PEC · PhD Mechanical

Big-data pipeline

★★★★★

I needed an ablation table I had not run myself, two days before submission. Their engineer turned it around in eighteen hours, with a methods paragraph in my voice. Reviewer 3 quoted the ablation favourably.

Sanjay Singh

JNU · M.Tech ML

Algorithm replication

Questions

FAQ

It is implementation of your method, against your equations, your data, and your supervisor's direction — and you sign off at every stage. Whether to acknowledge our engineer in the thesis is your decision, in line with your university's authorship policy. Most clients credit us in the acknowledgements the same way they would credit a lab technician or a software engineer; some do not. We will follow whatever you decide and we will not list ourselves on any thesis or paper without your explicit consent.

§ Begin

Send the method chapter. Receive a runnable repo, on time, in your notation.

Twelve engineer-editors, paired with a PhD subject lead. Five fixed touchpoints. One named engineer on the file. Plan by Day 2, first runnable by Day 5, real numbers by Day 8, handoff by Day 10 — with README, walk-through video, methods note, and Dockerfile.

Turnaround · 5–10 days · longer for systems
Pricing · From ₹18,000
Guarantee · 5% credit per day late · runs on a fresh machine
Support · One engineer, named, from kick-off to handoff

New submission

id · DRAFT-26.04

Choose Service *

No. of Pages *

Document Language *

Document *

Drop file or click to upload

.docx · .doc · .pdf · max 50 MB

Email ID *

First name *

Last name

Phone (with country code) *

Country *

Receive OTP on

SMS Email WhatsApp (coming soon)

I have read the Privacy Policy and agree to the terms.

Note: Due to a delay in downloading reports from Turnitin, some report deliveries may be delayed. We appreciate your patience.

Your method, made runnable — implemented by an engineer who has read your draft.

New submission

What arrives in your inbox — a runnable repo, not a folder of scripts.

What we implement, by stack and by what your chapter actually is.

Machine learning · NLP

Statistics · biostatistics · econometrics

Image, signal & medical processing

Networking · simulation · MANET

Big data & backend

One engineer, one subject lead. Five fixed touchpoints. No chase-up.

Kick-off call

Implementation plan

First runnable

Numbers + ablations

Repo handoff + walk-through

Brief to handoff — step by step.

Brief upload

Quote · within five minutes

Engineering coordination

Repo handoff

By chapter shape, not by line count.

It runs on a fresh machine. Or we have not finished.

Scholars whose thesis code we wrote — before the service paused, and since reopening.

FAQ

Send the method chapter. Receive a runnable repo, on time, in your notation.

New submission