>S//
●loading ideas...
S// LABSVERIFIED
TICKET NO. 2026 // ADMIT ONE

Cut across the line➔

[ Hold left click & drag scissor across dashed line to enter ]

Sandesh Anvekar // Founder Mode PortfolioPERFORATED ADMISSION • 2026
S// LABS · PROTOTYPING LIVEPUNE, IN

WELCOME TO THEBUILD LABOF SANDESH.

AI PRODUCTS·AGENTS·AUTOMATION·STARTUP EXPERIMENTS

“I turn ambitious ideas into AI-powered products, systems, and experiments people can actually use.”

EXPLORE BUILDS & DOSSIERS→
[ SYSTEM ARCHITECTURE LAB ]SCROLL TO ASSEMBLE CAPABILITIES ↓
01 // AUDIOVOICE AI (<450MS)
02 // AGENTSAUTONOMOUS AGENTS
03 // DATARAG & VECTOR ARCHITECTURE
04 // REASONINGGROQ & LLAMA 3.3
05 // SYSTEMSFULL-STACK PRODUCT SHIPPER
06 // CRAFTFRONTEND CREATIVE MASTERY
01 // AUDIO
VOICE AI
(<450MS)
02 // AGENTS
AUTONOMOUS AGENTS
AUTONOMOUS AGENTS
03 // DATA
RAG &
VECTOR ARCHITECTURE
04 // REASONING
GROQ &
LLAMA 3.3
05 // SYSTEMS
FULL-STACK PRODUCT
SHIPPER
06 // CRAFT
FRONTEND CREATIVE
MASTERY
AI AGENTS✦
FULL-STACK PRODUCTS●
RAG SYSTEMS✦
VOICE AI●
AUTOMATION✦
STARTUP BUILDS●
MCP & TOOLS✦
HIGH-AGENCY EXECUTION●
AI AGENTS✦
FULL-STACK PRODUCTS●
RAG SYSTEMS✦
VOICE AI●
AUTOMATION✦
STARTUP BUILDS●
MCP & TOOLS✦
HIGH-AGENCY EXECUTION●
S// LABSVERIFIED
FOUNDER-MINDED AI BUILDER·PUNE, INDIA

WHO ISSANDESH?

[ SPECIMEN LANDING PAD ]
✦✦✦
Sandesh Anvekar Cartoon Hero
SKILLS CRUSHED ✦ MASTEREDSANDESH ANVEKAR
[ SPECIMEN LANDING PAD · S// LABS ]READY FOR DESCENT

I’m a CS + AI/ML student from Pune.

I build ambitious systems, test product ideas, and learn by shipping. My goal is not to collect tools or hoard certificates—it is to build things that matter and solve genuine human friction.

BUILDER AGE

19 // BUILDING EARLY

B.Tech CS + AI/ML @ NIAT Pune / ADYPU

Unusually high-agency early-career builder. Focused on practical product engineering rather than passive coursework.

VERIFIED PROOFAGE 19 // SHIP EARLY
COMPETITIVE SHIP
?

FOUNDER VOICE // FINALIST

OpenAI × NIAT Buildathon (Runner-Up)

Architected and shipped an adversarial AI pitch trainer under high-pressure constraints; selected as Finalist & Runner-Up in the OpenAI × NIAT Buildathon.

VERIFIED PROOFFINALIST ✦ RUNNER-UP
LEADERSHIP
SHIP IT

EDC PRESIDENT

Entrepreneurship & Development Club

Elected club president leading 100+ student founders, hosting developer roundtables, and driving campus buildathon culture.

VERIFIED PROOF100+ BUILDERS
SYSTEM CRAFT

AI + SOFTWARE + STARTUPS

Full-Stack Systems & Venture Mindset

Passionate about building self-sustaining software, automated workflows at Vectr Labs, and sub-second voice agent interfaces.

VERIFIED PROOFSYS // LOW LATENCY
FLAGSHIP BUILD // 01
🏆OPENAI × NIAT BUILDATHON · RUNNER-UP

FOUNDERVOICE.

An adversarial AI investor that pressure-tests startup pitches in sub-400ms voice dialogue before the real room does.

SPECS:LATENCY: <400MS P95SPEECH: DEEPGRAM NOVA-2REASONING: GROQ LLAMA 3.3 70B
SYSTEM SPEC // CAPABILITIES
4 MODULES
01 //< 400MS P95
✦ SUB-400MS VOICE LOOP

Zero-stall bidirectional speech loop with instant interruptions.

02 //ADVERSARIAL
✦ ZERO-SYCOPHANCY VC

Probes unit economics, CAC/LTV disconnects, and defensibility.

03 //REALTIME INGEST
✦ MULTIMODAL SLIDE OCR

Deep-scans slides & financial sheets before opening dialogue.

04 //4-PHASE TRIAL
✦ SCENE-BY-SCENE SIM

Live partner room crucible with forensic audit scorecards.

INTERACTIVE TESTBENCH:

HOW THE PROJECT WORKS // THE FOUNDER EXPERIENCE

✦ ZERO-SYCOPHANCY✦ OPENAI × NIAT RUNNER-UP
01Deck Ingest
✦ Drop Deck & Financials

Upload slides PDF and spreadsheet. Pipeline deep-scans unit economics, CAC/LTV, and assumptions before a single word is spoken.

[ DOCUMENT PARSER ]

Parsed 'Series_A.pdf' (14 slides) + 'Model.xlsx'

▸ Discrepancy: $4.2M ARR claimed vs 4.1% monthly churn.

● 14 SLIDES PARSED
02Voice Pitch
✦ Spoken Dialogue

Speak naturally into your mic as if seated in a partner room. Pure conversational flow with sub-400ms audio turnaround.

[ FOUNDER (YOU) ]

“We build the autonomous layer for engineering teams...”

▸ Sub-400ms speech stream cross-checked against slide data.

● SPEECH LOOP ACTIVE
03Cross-Exam
✦ Live Interruptions

AI partner cuts in mid-sentence to pressure-test shaky metrics, timeline contradictions, and defuse weak defenses.

[ AI INVESTOR (INTERRUPTING) ]

“Stop there. Sales cycle is 2 weeks, but slide 8 shows 6-mo bank pilots?”

▸ Tactical interrupt: Narrative vs pilot timeline conflict.

● HIGH PRESSURE
04Scorecard
94%
✦ Runway & Fixes

Unsparing forensic audit: Runway Survival Probability, defensibility vulnerabilities, and concrete narrative fixes.

[ POST-PITCH AUDIT ]

Runway Survival: 78% (Up from 42% on baseline run)

▸ Fix Slide 11: Switch TAM to bottom-up ACV math ($120k x 450 logos).

● DEFENSIBILITY READY
FINAL OUTCOME // COMPLETE PITCH READINESS:Founders enter partner meetings knowing their weakest metrics, rehearsed under intense live fire.
✦ SUB-400MS ENGINE
STAGE // LAUNCH
✦ INTERACTIVE FILM
watch a simulation[ Project Founder Voice ]
[ HIT THE RED BUTTON TO LAUNCH ]
PRESS TO EXPERIENCE THE SUB-400MS ADVERSARIAL CROSS-EXAM
OPEN COMPLETE BUILD DOSSIER→
S// LABSVERIFIED

SELECTED BUILD DOSSIERS.

A physical spiral doodle sketchbook of shipped products, operational automation engines, and active lab prototypes. Flip through the pages below.

DOSSIER // 01[AI PRODUCT]
SHIPPED & TESTEDPG. 1
✦ AI INVESTOR CROSS-EXAMINATION:

“Where's your distribution moat? Polite friends won't grill you, but I will.”

LOW-LATENCY VOICE ENGINE
< 450MS LOOP
GROQ (LLAMA 3.3 70B)·DEEPGRAM NOVA-2
TAP TO SWITCH VC PROMPT→SIDE A // ADVERSARIAL
interactive prototype · tap or hover above
★ OpenAI × NIAT Buildathon Finalist & Runner-Up

FOUNDER VOICE

An AI investor that pressure-tests startup pitches before the real room does.

BUILDER INTENT // WHY THIS EXISTS:

Most pitch practice happens with polite colleagues who give friendly head-nods. In real VC meetings, investors grill unit economics and moats. I architected this low-latency voice AI to pressure-test founders before the real room does.

TECHNOLOGY STACK:
Next.jsReactTypeScriptTailwind CSSSupabase (PostgreSQL + RLS)
ROLE:

System Architecture, Voice AI Pipeline, Full-Stack Build & Prompt Engineering

OPEN FULL DOSSIER→
SKETCHBOOK ENTRY #1
S// LABSVERIFIED

FROM “COOL IDEA”TO“IT ACTUALLY WORKS.”

The engineering discipline behind high-agency shipping: execution over endless research, and reality over tool-switching.

WASHI // 01
📌RULE #01SCRAPBOOK

“Stop overthinking. Build the smallest prototype that answers the real question.”

sandesh’s sketchpad ✎★
TAPED // LAB
RUN SHEET // REALITYSHIP-READY
  • ✓Talk to real humans directly
  • ✓48-hour prototype sprint
  • ✓Break assumptions in code
  • ✗Endless roadmap meetingsKILLED ✂
3 AM ESPRESSO
SHIP IT.
++++
EXPEDITION TRACK //STAGE 01 OF 04 DOCKED
++++
STAGE 01 //ROUGH SKETCH // QUESTION ASSUMPTIONS

FIND THE REAL PROBLEM

Filter out vanity toys & superficial wrappers

WORKING PRINCIPLE:

“An agent is not always the answer.”

Before writing an LLM prompt or orchestrating a workflow, identify whether the user friction is high-stakes. If deterministic logic or a simple SQL query solves the issue faster and cheaper, don't force AI into it.

INTERACTIVE PROVING GROUNDACTIVE EXPEDITION DOCKED
DIAGNOSTIC 01 // HYPE DECIMATORVERIFIED OPTIMAL
SQL / LOGIC5-AGENT LOOP
EST. LATENCY4.2 ms
RELIABILITY99.99%
++++
STAGE 02 //WIREFRAME // MINIMAL VIABLE ARCHITECTURE

BUILD THE SMALLEST USEFUL SYSTEM

Laser-focused 48-hour prototype scope

WORKING PRINCIPLE:

“A prototype should answer a question.”

Every prototype is an experiment designed to resolve one core risk: Can the model judge pitches objectively? Can we stream audio under 400ms? Build only what is needed to prove or kill the premise.

INTERACTIVE PROVING GROUND
DIAGNOSTIC 02 // 48-HOUR SCOPE CALIPERSPEED: 48 HOURS
TEST OBJECTIVE:1 Question: Audio <400ms?
VERDICT:Kill or prove premise now
++++
STAGE 03 //CRAFT // VISUAL FEEDBACK & ERGONOMICS

MAKE IT FEEL REAL

Tactile interface, clear feedback, zero clunkiness

WORKING PRINCIPLE:

“Good UI is part of the product, not decoration.”

A brilliant backend with a laggy, confusing UI feels like broken software. Latency indicators, optimistic updates, and instant visual feedback turn a raw script into an experience people trust.

INTERACTIVE PROVING GROUND
DIAGNOSTIC 03 // TACTILITY TACHOMETERRESPONSE: 14 MS
Clicks Recorded: 0
UI STATESTANDBY IDLE
USER PERCEPTIONZero Perceived Friction
++++
STAGE 04 //PRODUCTION // ITERATION CADENCE

TEST. LEARN. SHIP AGAIN.

Put it in front of ruthless users immediately

WORKING PRINCIPLE:

“Shipping teaches faster than planning.”

Theoretical roadmaps cannot predict where real humans stumble. We ship to buildathons, campus clubs, and real users to let reality prune our assumptions, then iterate immediately.

INTERACTIVE PROVING GROUND
DIAGNOSTIC 04 // ITERATION RADARRELEASE CADENCE: v1.1
FEEDBACK ARENA:Ruthless Real Users
ASSUMPTIONS PRUNED:3 Iterations Tested
S// LABSVERIFIED

CURRENTLYTINKERING WITH.

Six active R&D focus areas on Sandesh's workbench. No meaningless logo walls or fake percentage bars—just honest technical questions being explored through real code.

WORKBENCH INSPECTION // 01ACTIVE EXPLORATION

AGENTS & ORCHESTRATION

Autonomous execution with verifiable guardrails

Chaining LLMs with tools, dynamic planners, and self-correcting state machines instead of brittle single-prompt scripts.

THE HONEST LESSON:

“An agent is not always the answer. Most production workflows only need deterministic code for 80% of steps and an LLM for the remaining 20% ambiguous reasoning.”

ACTIVE MECHANISMS:
Tool CallingMCP ServersState GraphsSubagent Hand-offs
S// LABSVERIFIED

I DON’T ONLYBUILD SOFTWARE.

“I care about the customer, the pitch, the team, the distribution, and whether the idea deserves to exist.”

LEAD PHILOSOPHY // EVIDENCE DOSSIER

Culture, Club Leadership & Execution Velocity

As President of the Entrepreneurship & Development Club (EDC), I lead teams to run buildathons, host founder roundtables, and foster a culture of active shipping rather than passive theory.

Push Pin
LEADERSHIP // 01COMMUNITY & LEADERSHIP

PRESIDENT, EDC

Entrepreneurship & Development Club @ ADYPU

Leading initiatives, organizing hands-on hackathons, mentoring peers on product ideation, and transforming a student club into an active early builder community.

CASE DOC // 01VERIFIED
PROOF:100+ Active Builders Reached
✦ GRAB & PIN #1
Push Pin
COLLABORATION // 02COMPETITIVE SHIP

BUILDATHON POD CAPTAIN

OpenAI × NIAT Buildathon

Rallied a cross-disciplinary team under high-pressure build constraints to architect and ship Founder Voice, securing the Finalist (Runner-Up) honor in the OpenAI × NIAT Buildathon.

CASE DOC // 02VERIFIED
PROOF:Finalist & Runner-Up
Push Pin
DISPATCH // 03NETWORKING

FOUNDER ROUNDTABLES

Bringing Real Founders to Campus

Curated intimate conversations between active technology entrepreneurs and student builders, focusing on unvarnished truths of PMF, bootstrapping, and venture capital.

CASE DOC // 03VERIFIED
PROOF:Direct Founder Dialogue

NOTE FROM LAB:

“Execution without ego. Iterate in public.”

EDC PUNE // VECTR LABS AUTOMATION
SELECT TABS ABOVE TO SWITCH PILLARS
S// LABSVERIFIED

CURRENT SIGNAL // NOW

LIVE SIGNAL DISPATCH
OCTOBER 2026
NOTEBOOK ENTRY // 01 · REAL-TIME PRESENCE● BUILDING
Current Location & Timezone:
📍 PUNE, MAHARASHTRA, INDIA
PUNE LOCAL CLOCK (IST):02:40:00 AM IST
INDEX CARD // ACADEMIC RECORD & LABOFFICIAL RECORD
Institution & Degree:

NIAT Pune / ADYPU — B.Tech Computer Science (AI/ML)

Academic Performance:
Cumulative CGPA: ~8.95 / 10.0
Campus Leadership:

🏛️ President, Entrepreneurship & Development Club (EDC)

★ ACCREDITED
POST-IT // CORE OBSESSION

Voice-first AI interactions, autonomous multi-agent systems, and scalable product architecture.

#HIGH_AGENCY#EXECUTION_FIRST
KRAFT SCRAP // ACTIVE LAB BENCH EXPLORATIONS
  • ☑Ultra-low latency speech-to-speech pipelines (< 400ms end-to-end)
  • ☑Standardizing local tool calling workflows using Model Context Protocol (MCP)
  • ☑Adversarial evaluation engines for LLM applications
  • ☑Product-led growth loops for developer tooling & early SaaS
FOUNDER CONVICTION // RAW THINKING
“The biggest moat in AI right now isn't access to weights—it's how fast you turn raw model intelligence into an intuitive product people use every day without feeling like they are talking to a chatbot.”
SANDESH // S_LABS
HOVER OVER STICKERS TO PEEL:
VOICE AI (<400MS)
HIGH-AGENCY SHIPPER
DOODLE LAB VERIFIED
S// LABSVERIFIED
[OPEN TO COLLABORATION]

GOT A PROBLEMWORTH BUILDING?

I’m interested in ambitious projects, strong internships, sharp collaborators, and difficult technical problems.

S// LABSVERIFIED
S// LABS · BUILT BY SANDESH ANVEKAR
PUNE, INDIA // ALL SYSTEMS TESTED & VERIFIED