Work

Eight case studies: seven AI systems I built at Masterly Solutions and one personal project. Each covers the problem, the design decisions and the results, with evidence for every number.

Illustration
What do I do if the cash drawer is short at closing?
Recount with a second staff member present, then record the difference on the closing sheet.
Cash Handling SOP, page 14
If the closing count is short, recount with a second staff member present and record the variance on the closing sheet.
Staff ask in plain English
No digging through hundreds of pages of procedures.
The answer comes from company documents
The AI answers from the passages it finds, not from memory.
Every answer shows its source
One click opens the exact page, with the passage highlighted.
DocuPilotEnterprise RAG PlatformCut procedure questions to supervisors by an estimated ~40%, with a cited source page for every answerragdoclingfaiss
Illustration
When does my team play next?
Field 4, tomorrow at 2:30 PM. Check in at the field desk 45 minutes before.From your team's schedule
Can we bring our own canopy to the field?
Staff correction, saved to the knowledge baseCanopies are allowed behind the dugout fence, not along the foul lines.
Yes, team canopies are allowed behind the dugout fence, but not along the foul lines.From a staff correction
Personal answers
It reads the family's own schedule, not just the general FAQs.
Staff fix a wrong answer once
The correction goes straight into the knowledge base.
Everyone gets the fixed answer
The next person who asks gets the corrected answer, so fewer questions reach staff.
RAG Support ChatbotCustomer Support AssistantHandled 30,000+ questions for 8,800+ people, with staff corrections fed straight back into its knowledge baseraghitlfaiss
Illustration
Analyst's ideaHome team's win rate at this ground over the last two years
def home_win_rate(match): games = past_matches(match.ground, team=match.home, years=2) return wins(games, match.home) / len(games) if games else 0.5
Matches the worked examples
Same answer on every run
Never sees future results
Flips correctly when teams swap
Approve feature
An analyst describes the idea
In plain English, with no code.
An AI agent writes the code
It turns the idea into a prediction feature.
Automatic tests catch bad code
If a test fails, the agent repairs the code and tries again.
A person approves it
Nothing goes live without a human sign-off.
CrickPal Intelligence LayerAI Win-Prediction PlatformAnalysts turn plain-English ideas into tested prediction features, without waiting on an engineeragentsevalsml
Illustration
Northern Hawks132 for 4after 16 of 20 overs, against Coastal Kings
Hawks recover from 42 for 3 with a 90-run stand
Early wickets left the Hawks in trouble, but a patient partnership turned the innings around. They head into the last four overs with wickets in hand and a strong total in sight.
Over 6Over 12Over 16, latest
Live match data comes in
Every ball's score streams in as the match is played.
The AI writes the story
A headline and short summary fans can read in seconds.
It keeps up with the match
Refreshed as play goes on, on a self-hosted model with no per-token bill.
CrickPal AI SummaryReal-Time LLM Commentary EngineLive match summaries for fans every few deliveries, on a self-hosted LLM with no per-token costollamareal-timefastapi
Illustration
Player 7, 91% match
Player 12, 88% match
62%, needs a check
2 playerstagged automatically
1 facesent to staff to check
Faces matched to players
Each face in a tournament photo is compared with player photos on file.
Sure matches tag themselves
Unsure ones go to a person to check. Manual tagging dropped by an estimated ~70%.
AI Face Matching SystemGPU Photo-Tagging PipelineCut manual photo tagging by an estimated ~70%, with players tagged automatically and unattendedpytorchfacenetcuda
Illustration
What's my EMI on a ₹20 lakh home loan at 9% over 15 years? And what if the rate drops?
Understood asHome loan₹20,00,00015 years9.0%, 8.5%, 8.0%
Rate Monthly EMI Total interest
9.0% ₹20,285 ₹16,51,360
8.5% ₹19,695 ₹15,45,062
8.0% ₹19,113 ₹14,40,348
Borrowers ask in plain English
About any of six loan products.
The AI turns it into exact inputs
Loan type, amount, term and the rates to compare.
A calculation engine does the maths
Every figure is calculated, never made up by the AI.
QuantumCopilotAI Loan & Financial AssistantBorrowers compare loan options in plain English, with every number from a calculation engine, not the LLMllmasynciofastapi
Illustration
Anywhere good to eat near the fields?
Three family places are within a 10-minute walk, and two take large groups without a booking.Late-night foodParking nearby
Can you book us a table for tonight?
I can't make bookings, but the event help desk can point you to places that take reservations.
Answers from about 50 FAQs
They all fit in one cached prompt, so there's no search step that can miss.
Suggests what to ask next
Only from the real FAQ list, never invented.
Clear limits
Out-of-scope requests get a fixed reply, so it never promises anything.
Concierge ChatbotLong-Context Visitor AssistantStarted as a RAG pipeline, rebuilt as one cached prompt with no retrieval to go wronglong-contextprompt-cachingllm
Illustration
Lists, beginner quiz
What does len([3, 1, 4]) return?3 ✓
Which method adds an item to the end of a list?append() ✓
For nums = [2, 4, 6], what is nums[-1]?2 ✗ key: 6
What does [1, 2] + [3] give?[1, 2, 3] ✓
3 of 4 correct, 75%. Pass mark 70%Next topic unlocked: Dictionaries
AI agents write the lesson and quiz
Every question comes with its answer key.
Code grades the answers
Checked against the key, so the same answers always get the same score.
Pass to move on
70% on a topic quiz unlocks the next topic.
ViperMindMulti-Agent Python TutorA Python tutor where LLM agents teach and code grades, so the same answers always get the same resultlanggraphredisdocker